feat(plip): voice loop + DTMF + ring cadence
CI / platformio (pull_request) Failing after 4m0s
CI / platformio (push) Failing after 14m58s

- hook polarity active-HIGH + auto-resync (was LOW)
- ring cadence FT 1.5s ON / 3.5s OFF
- DTMF Goertzel decoder (dtmf.c/h) + rotary debounce
- LISTEN half-duplex: capture → /v1/voice/reply → play
- WAV playback buffered PSRAM + mono→stereo upmix
- SPIFFS mount at boot for pre-loaded greetings
- ES8388: DAC digital vol + mic PGA + GPIO INPUT_OUTPUT
- turn_client multipart + 90s timeout + fixed routing
- debug endpoints: vol/dacvol/offhook/getfile/hookmon
This commit was merged in pull request #25.
This commit is contained in:
clement
2026-06-15 21:12:33 +02:00
parent 37db47ad7b
commit aa7ae277ed
16 changed files with 1263 additions and 114 deletions
+1
View File
@@ -10,6 +10,7 @@ idf_component_register(
"tones.c"
"dialer.c"
"conversation.c"
"dtmf.c"
"turn_client.c"
"slic.c"
INCLUDE_DIRS "."
+31 -1
View File
@@ -31,7 +31,7 @@ menu "PLIP Voice Configuration"
config PLIP_SPEAKER_VOLUME
int "Default Speaker Volume (0-100)"
default 70
default 80
range 0 100
help
Default speaker output volume at boot.
@@ -71,6 +71,18 @@ menu "PLIP Voice Configuration"
a pulse train, the train is considered complete and the digit is
emitted. 200 ms is standard for French rotary dials.
config PLIP_DIAL_DTMF
bool "Enable DTMF (touch-tone) dialing via Goertzel"
default n
help
When enabled, a background task reads 20 ms microphone frames and
runs a Goertzel-based DTMF detector (8 frequencies: 697-1633 Hz).
Confirmed digits (≥ 40 ms tone, with twist and dominance guards)
are pushed to the dialer just like rotary pulses.
The detector is active only between off-hook and the start of the
NPC greeting; it is disarmed during voice capture (CONNECTED state).
Can be combined with PLIP_DIAL_PULSE: whichever source detects a
digit first wins. Default off — enable for touch-tone handsets.
config PLIP_GATEWAY_URL
string "NPC Gateway Base URL"
@@ -88,4 +100,22 @@ menu "PLIP Voice Configuration"
Bearer token sent as "Authorization: Bearer <token>" on every
/v1/voice/turn request. Leave empty to skip the header.
config PLIP_VOICE_REPLY
bool "Enable Stage-3 conversational LISTEN loop (capture -> /v1/voice/reply -> play)"
default n
help
When enabled, after the NPC greeting is played (STATE_CONNECTED),
the firmware enters a continuous listen loop:
1. Capture mic audio (up to 8 s, VAD-gated) via audio_capture_wav().
2. POST the captured WAV as multipart/form-data to
CONFIG_PLIP_GATEWAY_URL/v1/voice/reply (STT + NPC reply via Kyutai).
3. Play the NPC response WAV from /spiffs/reply.wav.
4. Repeat until the handset is hung up.
Requires the gateway (zacus-gateway FastAPI) to be reachable and
the /v1/voice/reply endpoint to be operational.
Capture buffer (~256 KB for 8 s) is allocated from PSRAM when
available; falls back to internal heap with reduced duration (4 s).
Leave OFF (default) to keep STATE_CONNECTED as a terminal state
(Stage 2 behaviour — greeting only, no further interaction).
endmenu
+74 -62
View File
@@ -69,6 +69,7 @@ static QueueHandle_t s_queue;
static i2s_chan_handle_t s_spk_handle = NULL;
static i2s_chan_handle_t s_mic_handle = NULL;
static volatile bool s_stop_req = false;
static volatile bool s_playing = false; /* true while the worker plays a clip */
static bool s_sd_mounted = false;
static bool s_spiffs_mounted = false;
@@ -81,7 +82,7 @@ static void ensure_spiffs_audio(void)
.base_path = "/spiffs",
.partition_label = "storage",
.max_files = 8,
.format_if_mount_failed = false,
.format_if_mount_failed = true, /* format a blank/corrupt partition at boot */
};
esp_err_t ret = esp_vfs_spiffs_register(&conf);
if (ret == ESP_OK || ret == ESP_ERR_INVALID_STATE /* already mounted */) {
@@ -167,45 +168,6 @@ static esp_err_t parse_wav_header(const uint8_t *buf, size_t len, wav_info_t *ou
return ESP_OK;
}
/* ── Streaming helpers ───────────────────────────────────────────────────── */
/* Stream raw PCM-16 data to the speaker I2S channel in chunks. */
static void stream_pcm(const uint8_t *data, size_t byte_len)
{
size_t offset = 0;
while (!s_stop_req && offset < byte_len) {
size_t chunk = (byte_len - offset < 2048) ? (byte_len - offset) : 2048;
size_t written = 0;
esp_err_t ret = i2s_channel_write(s_spk_handle,
data + offset, chunk,
&written, pdMS_TO_TICKS(500));
if (ret != ESP_OK) {
ESP_LOGW(TAG, "I2S write error: %s", esp_err_to_name(ret));
break;
}
offset += chunk;
}
}
/* Play WAV from an in-memory buffer. */
static void play_wav_buf(const uint8_t *buf, size_t len)
{
wav_info_t wi = {0};
esp_err_t ret = parse_wav_header(buf, len, &wi);
if (ret != ESP_OK) {
ESP_LOGW(TAG, "WAV parse error: %s", esp_err_to_name(ret));
audio_play_tone(880.0f, 200);
return;
}
if (wi.bits_per_sample != 16) {
ESP_LOGW(TAG, "WAV: %d-bit not supported (need 16-bit)", wi.bits_per_sample);
return;
}
ESP_LOGI(TAG, "WAV: %"PRIu32" Hz %d-bit %d ch, %"PRIu32" bytes PCM",
wi.sample_rate, wi.bits_per_sample, wi.channels, wi.data_size);
stream_pcm(buf + wi.data_offset, wi.data_size);
}
/* WAV streaming chunk size — keeps heap usage well under 8 KB. */
#define PLAY_CHUNK_BYTES 4096
@@ -279,28 +241,62 @@ static void play_wav_file(const char *path)
return;
}
/* Stream PCM to I2S in small chunks — no large malloc needed. */
static uint8_t s_play_chunk[PLAY_CHUNK_BYTES]; /* static: avoids stack pressure */
uint32_t remaining = wi.data_size;
size_t total_written = 0;
/* The I2S TX slot is STEREO @ SAMPLE_RATE. A mono WAV must be expanded to
* L+R or it is consumed at 2x rate (the "chipmunk" fast/high-pitch bug).
* A WAV whose rate differs from SAMPLE_RATE would also play at the wrong
* speed — warn (the gateway TTS always returns 16 kHz, matching). */
if (wi.sample_rate != SAMPLE_RATE) {
ESP_LOGW(TAG, "WAV rate %"PRIu32" Hz != I2S %d Hz — playback speed will be off",
wi.sample_rate, SAMPLE_RATE);
}
const bool mono = (wi.channels == 1);
while (!s_stop_req && remaining > 0) {
uint32_t to_read = (remaining < PLAY_CHUNK_BYTES) ? remaining : PLAY_CHUNK_BYTES;
size_t n = fread(s_play_chunk, 1, to_read, f);
if (n == 0) break;
size_t i2s_written = 0;
esp_err_t ret = i2s_channel_write(s_spk_handle, s_play_chunk, n,
&i2s_written, pdMS_TO_TICKS(500));
/* Load the ENTIRE PCM into PSRAM BEFORE playing. Streaming fread() from
* SPIFFS *between* I2S writes stalls the DMA → underrun → audible
* distortion ("saturation"). Diagnostic confirmed: the stored WAV is clean
* (0% clip) and RAM-generated tones play clean, but SPIFFS-streamed WAVs
* distorted. Reading it all up-front = zero file I/O during playback. */
uint8_t *pcm = heap_caps_malloc(wi.data_size, MALLOC_CAP_SPIRAM);
if (!pcm) pcm = malloc(wi.data_size);
if (!pcm) {
ESP_LOGE(TAG, "play: OOM for %"PRIu32"-byte PCM buffer", wi.data_size);
fclose(f);
return;
}
size_t pcm_len = fread(pcm, 1, wi.data_size, f);
fclose(f);
static int16_t s_stereo_chunk[PLAY_CHUNK_BYTES]; /* mono→stereo scratch */
size_t off = 0;
size_t total_written = 0;
const size_t step = mono ? (PLAY_CHUNK_BYTES / 2) : PLAY_CHUNK_BYTES;
while (!s_stop_req && off < pcm_len) {
size_t bytes = (pcm_len - off < step) ? (pcm_len - off) : step;
const uint8_t *out = pcm + off;
size_t out_len = bytes;
if (mono) {
/* Duplicate each 16-bit mono sample into L and R. */
size_t samples = bytes / 2;
const int16_t *src = (const int16_t *)(pcm + off);
for (size_t k = 0; k < samples; k++) {
s_stereo_chunk[2 * k] = src[k];
s_stereo_chunk[2 * k + 1] = src[k];
}
out = (const uint8_t *)s_stereo_chunk;
out_len = samples * 4; /* 2 channels × 2 bytes */
}
size_t w = 0;
esp_err_t ret = i2s_channel_write(s_spk_handle, out, out_len, &w, pdMS_TO_TICKS(500));
if (ret != ESP_OK) {
ESP_LOGW(TAG, "I2S write error: %s", esp_err_to_name(ret));
break;
}
total_written += i2s_written;
remaining -= (uint32_t)n;
total_written += w;
off += bytes;
}
fclose(f);
float dur = (float)total_written / (float)(wi.sample_rate * wi.channels * (wi.bits_per_sample / 8));
free(pcm);
float dur = (float)total_written / (float)(SAMPLE_RATE * 2 * 2); /* 16k stereo 16-bit */
ESP_LOGI(TAG, "play done: %zu bytes written, %.2fs", total_written, dur);
}
@@ -385,6 +381,7 @@ static void audio_worker_task(void *arg)
ESP_LOGI(TAG, "play ignored: on-hook (handset down)");
break;
}
s_playing = true;
const char *p = cmd.path;
if (!p || !*p || strncmp(p, "embedded:", 9) == 0) {
ESP_LOGI(TAG, "play: embedded cue");
@@ -393,6 +390,7 @@ static void audio_worker_task(void *arg)
ESP_LOGI(TAG, "play: %s", p);
play_wav_file(p);
}
s_playing = false;
break;
}
default:
@@ -409,6 +407,11 @@ void audio_pa_set(bool enable)
ESP_LOGI(TAG, "PA %s", enable ? "ON" : "OFF");
}
bool audio_is_playing(void)
{
return s_playing;
}
esp_err_t audio_init(void)
{
/* 1. ES8388 I2C init + register sequence. */
@@ -418,6 +421,10 @@ esp_err_t audio_init(void)
return ret;
}
/* Apply the configured output volume. es8388_init() leaves OUT2 at 0 dB
* (max) — too loud for a handset earpiece — so set it explicitly here. */
es8388_set_volume(CONFIG_PLIP_SPEAKER_VOLUME);
/* 2. Create I2S channels (TX = speaker, RX = mic — full-duplex pair). */
i2s_chan_config_t chan_cfg = I2S_CHANNEL_DEFAULT_CONFIG(PLIP_I2S_NUM,
I2S_ROLE_MASTER);
@@ -513,6 +520,10 @@ esp_err_t audio_init(void)
8192, NULL, 5, NULL, 0);
if (ok != pdPASS) return ESP_ERR_NO_MEM;
/* Mount SPIFFS at init (not lazily) so turn_client can WRITE the NPC WAV to
* /spiffs before any playback has triggered the lazy mount. */
ensure_spiffs_audio();
ESP_LOGI(TAG, "audio init OK (I2S TX ready, RX handle allocated, ES8388 live)");
return ESP_OK;
}
@@ -564,9 +575,11 @@ int audio_capture_wav(uint8_t *out, size_t out_max, int max_ms, int silence_ms)
return -1;
}
/* Enable RX while keeping TX running (TX drives MCLK/BCLK/WS for the codec). */
/* RX is already enabled at boot (full-duplex). Calling enable again returns
* ESP_ERR_INVALID_STATE — that's fine, it just means RX is already running.
* Only a genuinely different error is fatal. */
esp_err_t ret = i2s_channel_enable(s_mic_handle);
if (ret != ESP_OK) {
if (ret != ESP_OK && ret != ESP_ERR_INVALID_STATE) {
ESP_LOGE(TAG, "capture: i2s_channel_enable(RX): %s", esp_err_to_name(ret));
free(rx_buf);
return -1;
@@ -639,8 +652,9 @@ int audio_capture_wav(uint8_t *out, size_t out_max, int max_ms, int silence_ms)
total_frames = f + 1;
}
/* Disable RX; TX was never stopped. */
i2s_channel_disable(s_mic_handle);
/* Leave RX enabled (full-duplex, as at boot). Disabling it here would make
* the next capture's enable a no-op INVALID_STATE AND break other RX users
* (streaming capture / DTMF) that assume RX stays running. */
free(rx_buf);
if (pcm_written == 0) {
@@ -706,8 +720,6 @@ int audio_capture_begin(int max_ms, int silence_ms)
return 0;
}
/*
* Read one 20 ms frame from the mic, downmix stereo→mono.
* mono_out must hold n_samples (320) int16_t values.
+5
View File
@@ -39,6 +39,11 @@ esp_err_t audio_play_async(const char *path);
/* Stop current playback immediately. */
esp_err_t audio_stop(void);
/* True while the audio worker is playing a clip (tone/WAV). The conversation
* LISTEN loop polls this to stay half-duplex: never capture while playing
* (avoids the earpiece→mic feedback that saturated the line). */
bool audio_is_playing(void);
/* Start ring tone cadence (ON 1s / OFF 2s) at ~440 Hz. Continues until
* audio_stop() is called. Non-blocking — spawns an internal task. */
esp_err_t audio_ring_start(void);
+143 -4
View File
@@ -20,16 +20,38 @@
#include "dialer.h"
#include "tones.h"
#include "audio.h"
#include "phone.h"
#include "turn_client.h"
#if CONFIG_PLIP_DIAL_DTMF
#include "dtmf.h"
#endif
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "esp_log.h"
#include "esp_timer.h"
#include "esp_heap_caps.h"
#include <stdio.h>
#include <string.h>
#if CONFIG_PLIP_VOICE_REPLY
/* Capture buffer sizing.
* PSRAM target: 8 s of 16kHz mono S16 + 44-byte WAV header.
* 8 s × 16000 samples/s × 2 bytes = 256000 bytes + 44 = 256044 → round to 256 KB.
* Fallback (internal heap, 4 s max):
* 4 s × 16000 × 2 + 44 = 128044 → round to 128 KB.
*/
#define CAPTURE_MAX_PSRAM (256 * 1024)
#define CAPTURE_MAX_IRAM (128 * 1024)
#define CAPTURE_MAX_MS_PSRAM 8000
#define CAPTURE_MAX_MS_IRAM 4000
#define CAPTURE_SILENCE_MS 800
#define REPLY_POLL_MS 200 /* interval for checking hook during playback */
#define REPLY_PLAYBACK_EXTRA_MS 500 /* safety margin added to computed WAV duration */
#define BETWEEN_TURNS_MS 300 /* short pause between capture rounds */
#endif /* CONFIG_PLIP_VOICE_REPLY */
#define TAG "conversation"
/* Duration of ringback before picking up and fetching the greeting */
@@ -54,6 +76,11 @@ static int64_t s_ringback_start_us = 0;
/* Session ID for the current call (generated at ringback → greet transition) */
static char s_sid[32] = {0};
/* Dialed number LOCKED at routing time. The dialer can keep accumulating
* spurious rotary pulses (marginal hook contact) during the call, so we must
* NOT re-read dialer_current() for the greeting/reply — that polluted number
* would 404 at the gateway. Capture the clean routed number here once. */
static char s_number[16] = {0};
/* Known numbers: ringback when dialed */
static const char *KNOWN[] = {
@@ -74,6 +101,9 @@ static void go_idle(void)
audio_stop();
audio_pa_set(false);
dialer_reset();
#if CONFIG_PLIP_DIAL_DTMF
dtmf_stop();
#endif
s_state = STATE_IDLE;
ESP_LOGI(TAG, "-> IDLE");
}
@@ -101,6 +131,9 @@ static void conv_task(void *arg)
if (s_state == STATE_IDLE) {
dialer_reset();
tones_dialtone_start();
#if CONFIG_PLIP_DIAL_DTMF
dtmf_start();
#endif
s_state = STATE_DIALTONE;
ESP_LOGI(TAG, "off-hook -> DIALTONE");
}
@@ -136,6 +169,9 @@ static void conv_task(void *arg)
const char *num = dialer_current();
if (is_known(num)) {
ESP_LOGI(TAG, "route %s -> known (ringback)", num);
/* Lock the routed number now — the dialer may pick up
* spurious pulses later and we must keep posting "17". */
snprintf(s_number, sizeof(s_number), "%s", num);
tones_ringback_start();
s_ringback_start_us = esp_timer_get_time();
s_state = STATE_RINGBACK;
@@ -158,6 +194,10 @@ static void conv_task(void *arg)
if (elapsed_ms >= RINGBACK_GREET_MS) {
/* Stop ringback tone synchronously before fetching */
tones_stop();
#if CONFIG_PLIP_DIAL_DTMF
/* Disarm DTMF before entering voice-capture phase */
dtmf_stop();
#endif
/* Generate a session ID from timer ticks */
snprintf(s_sid, sizeof(s_sid), "%lld",
(long long)esp_timer_get_time());
@@ -174,7 +214,7 @@ static void conv_task(void *arg)
break;
}
/* Fetch greeting WAV from gateway and enqueue playback */
if (turn_client_greeting(s_sid, dialer_current(),
if (turn_client_greeting(s_sid, s_number,
"/spiffs/turn.wav")) {
audio_play_async("/spiffs/turn.wav");
} else {
@@ -185,10 +225,104 @@ static void conv_task(void *arg)
break;
case STATE_CONNECTED:
/* Stage 3 will add listen/speak loop here */
#if CONFIG_PLIP_VOICE_REPLY
/*
* Stage 3 — LISTEN loop.
*
* Allocate capture buffer once from PSRAM (preferred) or internal
* heap. Then loop: capture → POST reply → play → wait → repeat.
* Exit on any on-hook event. Buffer freed before leaving.
*/
{
/* --- Allocate capture buffer -------------------------------- */
uint8_t *cap_buf = NULL;
size_t cap_max = 0;
int cap_ms = 0;
cap_buf = heap_caps_malloc(CAPTURE_MAX_PSRAM, MALLOC_CAP_SPIRAM);
if (cap_buf) {
cap_max = CAPTURE_MAX_PSRAM;
cap_ms = CAPTURE_MAX_MS_PSRAM;
ESP_LOGI(TAG, "listen: cap_buf %zu B from PSRAM", cap_max);
} else {
cap_buf = malloc(CAPTURE_MAX_IRAM);
if (cap_buf) {
cap_max = CAPTURE_MAX_IRAM;
cap_ms = CAPTURE_MAX_MS_IRAM;
ESP_LOGW(TAG, "listen: PSRAM unavail, cap_buf %zu B from heap (max %d s)",
cap_max, cap_ms / 1000);
} else {
ESP_LOGE(TAG, "listen: cap_buf alloc failed — staying silent");
/* Remain in CONNECTED without looping */
if (!s_offhook) go_idle();
break;
}
}
/* --- LISTEN loop ------------------------------------------- */
ESP_LOGI(TAG, "listen: entering loop (max %d s / silence %d ms)",
cap_ms, CAPTURE_SILENCE_MS);
while (s_offhook) {
/* HALF-DUPLEX: a telephone handset couples the earpiece into
* the mic. Never capture while anything is playing, or the
* playback feeds back and the line saturates. Wait for the
* greeting/filler/reply to finish, then let the line settle. */
vTaskDelay(pdMS_TO_TICKS(250)); /* let a just-queued clip start */
while (s_offhook && audio_is_playing()) {
vTaskDelay(pdMS_TO_TICKS(50));
}
if (!s_offhook) break;
vTaskDelay(pdMS_TO_TICKS(200)); /* line settle after playback */
/* Capture player utterance — nothing is playing now. */
int n = audio_capture_wav(cap_buf, cap_max,
cap_ms, CAPTURE_SILENCE_MS);
if (!s_offhook) break; /* hung up during capture */
if (n <= 44) {
ESP_LOGD(TAG, "listen: no voice (n=%d)", n);
continue;
}
ESP_LOGI(TAG, "listen: captured %d bytes, posting to gateway", n);
/* Filler "un instant, je traite votre demande" plays while the
* reply is synthesised (the POST blocks for several seconds). */
audio_play_async("/spiffs/wait.wav");
esp_err_t ret = turn_client_reply(s_sid, s_number,
cap_buf, (size_t)n,
"/spiffs/reply.wav");
if (!s_offhook) break; /* hung up during HTTP round-trip */
if (ret != ESP_OK) {
ESP_LOGW(TAG, "listen: turn_client_reply failed (%s) — skipping",
esp_err_to_name(ret));
continue;
}
/* Let the filler finish before the reply (no overlap), then play
* the reply. The loop top waits for it to end before re-capturing. */
while (s_offhook && audio_is_playing()) {
vTaskDelay(pdMS_TO_TICKS(50));
}
if (!s_offhook) break;
audio_play_async("/spiffs/reply.wav");
}
/* --- Cleanup ----------------------------------------------- */
free(cap_buf);
ESP_LOGI(TAG, "listen: loop exited (offhook=%d)", (int)s_offhook);
if (!s_offhook) go_idle();
}
#else
/* Stage 3 disabled — STATE_CONNECTED is terminal */
if (!s_offhook) {
go_idle();
}
#endif /* CONFIG_PLIP_VOICE_REPLY */
break;
case STATE_BUSY:
@@ -205,9 +339,14 @@ void conversation_init(void)
s_state = STATE_IDLE;
s_offhook = false;
s_hook_changed = false;
/* Stack bumped to 6144: STATE_GREET calls esp_http_client (blocking HTTP
* + file I/O) which needs more stack than the baseline 3072. */
/* Stack: STATE_GREET needs 6144 (esp_http_client + file I/O).
* STATE_CONNECTED listen loop (Stage 3) adds turn_client_reply (~1100 B
* locals) + stat() call → bump to 8192 when Stage 3 is compiled in. */
#if CONFIG_PLIP_VOICE_REPLY
xTaskCreate(conv_task, "conv", 8192, NULL, 4, NULL);
#else
xTaskCreate(conv_task, "conv", 6144, NULL, 4, NULL);
#endif
ESP_LOGI(TAG, "conversation init");
}
+323
View File
@@ -0,0 +1,323 @@
/*
* dtmf.c — DTMF (touch-tone) detector using the Goertzel algorithm.
*
* Detection pipeline per 20 ms frame (320 samples @ 16 kHz):
* 1. Compute Goertzel power for 8 DTMF frequencies (4 low + 4 high groups).
* 2. Find the strongest frequency in each group (best_low, best_high).
* 3. Apply three guards:
* a) Absolute energy threshold — both must exceed DTMF_ENERGY_THRESH.
* b) Group dominance ratio — winner must be > DTMF_DOMINANT_RATIO×
* times the second-best in the same group.
* c) Twist guard — energy ratio (low/high) must be within
* [1/DTMF_TWIST_MAX, DTMF_TWIST_MAX].
* 4. Debounce:
* - Require DTMF_CONFIRM_FRAMES consecutive matching frames to emit.
* - Require DTMF_RELEASE_FRAMES of silence/mismatch before re-arming.
*
* Threshold rationale:
* DTMF_ENERGY_THRESH = 4000000
* Goertzel power is mean-squared × N². A -30 dBFS sine at 16-bit PCM
* (amplitude ≈ 1000 LSB) gives power ≈ (1000²/2) × 320² / 320 ≈ 1.6e8.
* -50 dBFS (amplitude ≈ 100) gives ≈ 1.6e6. We set the floor at 4e6
* (≈ -47 dBFS) to reject noise while allowing quiet handset microphones.
*
* DTMF_DOMINANT_RATIO = 4.0f (≈ 6 dB separation within a group)
* A real DTMF tone drives exactly one row and one column. If two
* frequencies in the same group are within 6 dB of each other, it is
* more likely noise or voice than a keypad press.
*
* DTMF_TWIST_MAX = 8.0f (≈ 9 dB)
* ITU-T Q.24 allows up to ±8 dB twist between low/high group. We use
* 8× power ratio which corresponds to ≈ 9 dB — slightly relaxed to
* accommodate the varied mic responses of vintage telephone handsets.
*
* DTMF_CONFIRM_FRAMES = 2 (2 × 20 ms = 40 ms minimum tone duration)
* ITU-T Q.24 specifies ≥ 40 ms tone duration for valid DTMF.
*
* DTMF_RELEASE_FRAMES = 1 (≥ 20 ms inter-digit silence required)
* Prevents a single sustained keypress from re-triggering.
*/
/* sdkconfig.h must be included before any CONFIG_* test so the preprocessor
* has the symbol defined when the #if guard below is evaluated. */
#include "sdkconfig.h"
#if CONFIG_PLIP_DIAL_DTMF
#include "dtmf.h"
#include "audio.h"
#include "dialer.h"
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "esp_log.h"
#include <math.h>
#include <string.h>
#include <stdbool.h>
#include <stdint.h>
#define TAG "dtmf"
/* -------------------------------------------------------------------------
* Detection thresholds (see rationale in file header)
* ---------------------------------------------------------------------- */
/* Minimum Goertzel power (mean-squared × N) for a frequency to be considered
* "present". Both the row and column candidate must exceed this. */
#define DTMF_ENERGY_THRESH 4000000LL
/* Minimum ratio of best/second-best power within a group. Below this the
* group is ambiguous (noise / voice) and the frame is rejected. */
#define DTMF_DOMINANT_RATIO 4.0f
/* Maximum ratio of low-group / high-group power (and its inverse).
* Exceeding this means one group is far stronger than expected for DTMF. */
#define DTMF_TWIST_MAX 8.0f
/* Consecutive frames required before a digit is reported. */
#define DTMF_CONFIRM_FRAMES 2
/* Frames of "no valid tone" required between two reports. */
#define DTMF_RELEASE_FRAMES 1
/* -------------------------------------------------------------------------
* DTMF frequency table
* ---------------------------------------------------------------------- */
#define NUM_LOW 4
#define NUM_HIGH 4
static const float LOW_FREQS[NUM_LOW] = { 697.0f, 770.0f, 852.0f, 941.0f };
static const float HIGH_FREQS[NUM_HIGH] = { 1209.0f, 1336.0f, 1477.0f, 1633.0f };
/* DTMF matrix: [low_idx][high_idx] → character.
* Column 3 (1633 Hz) maps to ABCD which are unused on standard phones → '\0'. */
static const char DTMF_MATRIX[NUM_LOW][NUM_HIGH] = {
{ '1', '2', '3', '\0' }, /* 697 Hz row */
{ '4', '5', '6', '\0' }, /* 770 Hz row */
{ '7', '8', '9', '\0' }, /* 852 Hz row */
{ '*', '0', '#', '\0' }, /* 941 Hz row */
};
/* -------------------------------------------------------------------------
* Goertzel coefficient cache (precomputed at first call)
* ---------------------------------------------------------------------- */
#define FS 16000 /* sample rate */
#define N_SAMP 320 /* frame size */
static bool s_coeff_ready = false;
static float s_low_coeff[NUM_LOW];
static float s_high_coeff[NUM_HIGH];
static void precompute_coeffs(void)
{
for (int i = 0; i < NUM_LOW; i++) {
float k = (float)N_SAMP * LOW_FREQS[i] / (float)FS;
s_low_coeff[i] = 2.0f * cosf(2.0f * (float)M_PI * k / (float)N_SAMP);
}
for (int i = 0; i < NUM_HIGH; i++) {
float k = (float)N_SAMP * HIGH_FREQS[i] / (float)FS;
s_high_coeff[i] = 2.0f * cosf(2.0f * (float)M_PI * k / (float)N_SAMP);
}
s_coeff_ready = true;
}
/* -------------------------------------------------------------------------
* Goertzel power for a single frequency
* Power = Q1² + Q2² Q1·Q2·coeff (unnormalised, proportional to amplitude²)
* ---------------------------------------------------------------------- */
static float goertzel_power(const int16_t *samples, int n, float coeff)
{
float q1 = 0.0f, q2 = 0.0f;
for (int i = 0; i < n; i++) {
float q0 = coeff * q1 - q2 + (float)samples[i];
q2 = q1;
q1 = q0;
}
return q1 * q1 + q2 * q2 - q1 * q2 * coeff;
}
/* -------------------------------------------------------------------------
* Public API: dtmf_detect_frame
* ---------------------------------------------------------------------- */
char dtmf_detect_frame(const int16_t *mono, int n)
{
if (n <= 0 || !mono) return '\0';
if (!s_coeff_ready) precompute_coeffs();
/* Compute Goertzel power for all 8 frequencies */
float low_pow[NUM_LOW], high_pow[NUM_HIGH];
for (int i = 0; i < NUM_LOW; i++)
low_pow[i] = goertzel_power(mono, n, s_low_coeff[i]);
for (int i = 0; i < NUM_HIGH; i++)
high_pow[i] = goertzel_power(mono, n, s_high_coeff[i]);
/* Find strongest in each group */
int best_low = 0, best_high = 0;
float max_low = low_pow[0], max_high = high_pow[0];
for (int i = 1; i < NUM_LOW; i++) {
if (low_pow[i] > max_low) { max_low = low_pow[i]; best_low = i; }
}
for (int i = 1; i < NUM_HIGH; i++) {
if (high_pow[i] > max_high) { max_high = high_pow[i]; best_high = i; }
}
/* Guard (a): absolute energy threshold */
if ((int64_t)max_low < DTMF_ENERGY_THRESH || (int64_t)max_high < DTMF_ENERGY_THRESH)
goto no_tone;
/* Guard (b): group dominance — find second-best in each group */
{
float second_low = 0.0f, second_high = 0.0f;
for (int i = 0; i < NUM_LOW; i++) {
if (i != best_low && low_pow[i] > second_low)
second_low = low_pow[i];
}
for (int i = 0; i < NUM_HIGH; i++) {
if (i != best_high && high_pow[i] > second_high)
second_high = high_pow[i];
}
/* If second-best is within DTMF_DOMINANT_RATIO of best, ambiguous */
if (second_low > 0.0f && max_low < DTMF_DOMINANT_RATIO * second_low)
goto no_tone;
if (second_high > 0.0f && max_high < DTMF_DOMINANT_RATIO * second_high)
goto no_tone;
}
/* Guard (c): twist — power ratio must be within [1/TWIST_MAX, TWIST_MAX] */
{
float ratio = max_low / max_high;
if (ratio > DTMF_TWIST_MAX || ratio < (1.0f / DTMF_TWIST_MAX))
goto no_tone;
}
/* --- Debounce state (static) --- */
{
static char s_candidate = '\0';
static int s_confirm_count = 0;
static int s_release_count = 0;
static bool s_armed = true; /* true = ready to report */
char sym = DTMF_MATRIX[best_low][best_high];
if (sym == '\0') goto no_tone; /* 1633 Hz column — ignored */
/* Reset release counter: we have a tone */
s_release_count = 0;
if (!s_armed) {
/* Waiting for silence/release before accepting next press */
return '\0';
}
if (sym == s_candidate) {
s_confirm_count++;
} else {
s_candidate = sym;
s_confirm_count = 1;
}
if (s_confirm_count >= DTMF_CONFIRM_FRAMES) {
/* Confirmed — report and disarm until release */
s_confirm_count = 0;
s_candidate = '\0';
s_armed = false;
return sym;
}
return '\0';
no_tone:
/* No valid tone detected: advance release counter */
; /* label must precede a statement */
s_confirm_count = 0;
s_candidate = '\0';
if (!s_armed) {
s_release_count++;
if (s_release_count >= DTMF_RELEASE_FRAMES) {
s_armed = true;
s_release_count = 0;
}
}
return '\0';
}
}
/* -------------------------------------------------------------------------
* Background capture task
* ---------------------------------------------------------------------- */
#define DTMF_TASK_STACK 4096
#define DTMF_TASK_PRIO 3
#define DTMF_FRAME_SIZE 320
static volatile bool s_armed_flag = false; /* true = task should process frames */
static TaskHandle_t s_task_handle = NULL;
static void dtmf_task(void *arg)
{
(void)arg;
int16_t frame[DTMF_FRAME_SIZE];
int64_t rms_sq;
ESP_LOGI(TAG, "dtmf_task started");
/* Open the capture stream once and keep it open.
* Full-duplex: TX (speaker) stays active; RX is already enabled at boot.
* We pass generous max_ms / silence_ms since we never call capture_end
* while the call is in progress — dtmf_stop() just clears the armed flag. */
if (audio_capture_begin(3600000, 3600000) != 0) {
ESP_LOGE(TAG, "dtmf_task: capture_begin failed — task exits");
s_task_handle = NULL;
vTaskDelete(NULL);
return;
}
for (;;) {
if (!s_armed_flag) {
/* Disarmed: drain frames slowly so the RX FIFO doesn't overflow */
audio_capture_read_frame(frame, DTMF_FRAME_SIZE, &rms_sq);
vTaskDelay(pdMS_TO_TICKS(20));
continue;
}
int got = audio_capture_read_frame(frame, DTMF_FRAME_SIZE, &rms_sq);
if (got <= 0) continue;
char sym = dtmf_detect_frame(frame, got);
if (sym == '\0') continue;
ESP_LOGI(TAG, "DTMF detected: '%c'", sym);
if (sym >= '0' && sym <= '9') {
dialer_push_digit(sym - '0');
}
/* '*' and '#' are logged only — no dialer push for now */
}
}
void dtmf_start(void)
{
if (!s_task_handle) {
/* Create the task once */
BaseType_t ok = xTaskCreatePinnedToCore(
dtmf_task, "dtmf", DTMF_TASK_STACK, NULL, DTMF_TASK_PRIO,
&s_task_handle, 1);
if (ok != pdPASS) {
ESP_LOGE(TAG, "dtmf_start: xTaskCreate failed");
return;
}
}
s_armed_flag = true;
ESP_LOGI(TAG, "dtmf_start: DTMF detection armed");
}
void dtmf_stop(void)
{
s_armed_flag = false;
ESP_LOGI(TAG, "dtmf_stop: DTMF detection disarmed");
}
#endif /* CONFIG_PLIP_DIAL_DTMF */
+57
View File
@@ -0,0 +1,57 @@
#pragma once
/*
* dtmf.h — DTMF (touch-tone) detector via Goertzel algorithm.
*
* API:
* dtmf_detect_frame() — pure signal processing, no I/O, testable standalone
* dtmf_start() / dtmf_stop() — arm/disarm the background capture task
*
* The capture task reads 20 ms frames from audio_capture_read_frame() and calls
* dialer_push_digit() for confirmed digit presses (digits 0-9 only).
*
* Guard: all declarations and the task body are compiled only when
* CONFIG_PLIP_DIAL_DTMF is set (see Kconfig.projbuild).
*/
#include "sdkconfig.h"
#if CONFIG_PLIP_DIAL_DTMF
#include <stdint.h>
#ifdef __cplusplus
extern "C" {
#endif
/*
* Analyse one 20 ms frame (320 samples at 16 kHz) for a DTMF tone.
*
* Returns the detected character ('0'-'9', '*', '#') once per confirmed press,
* or '\0' when nothing is detected or debounce is still pending.
*
* Debounce rules (internal static state):
* - A symbol must appear in ≥ DTMF_CONFIRM_FRAMES consecutive frames to be
* reported.
* - After a report, at least DTMF_RELEASE_FRAMES of "no tone" must be seen
* before the same (or another) symbol can be reported again.
*/
char dtmf_detect_frame(const int16_t *mono, int n);
/*
* Start the DTMF background task (created once; idempotent re-arm).
* The task reads microphone frames and pushes confirmed 0-9 digits to the
* dialer. Must be called after audio_init().
*/
void dtmf_start(void);
/*
* Disarm the DTMF task. The task suspends itself; the RX stream continues
* for the benefit of the voice capture path (full-duplex architecture).
*/
void dtmf_stop(void);
#ifdef __cplusplus
}
#endif
#endif /* CONFIG_PLIP_DIAL_DTMF */
+25 -7
View File
@@ -151,7 +151,7 @@ esp_err_t es8388_init(void)
* - ADCCONTROL4 (0x0C) = 0x0C: I2S Philips 16-bit word length
* - ADCCONTROL5 (0x0D) = 0x02: ADCFsMode SINGLE SPEED RATIO=256 (16kHz@MCLK 4.096MHz)
* - ADCCONTROL8/9 (0x10/0x11) = 0x00: ADC digital volume 0dB */
if (i2c_write_reg(ES8388_ADC_CTL1, 0xBB) != ESP_OK) return ESP_FAIL; /* PGA +24dB L+R */
if (i2c_write_reg(ES8388_ADC_CTL1, 0x44) != ESP_OK) return ESP_FAIL; /* MIC PGA +12dB L+R (was +24dB: too hot for a close handset mic → clipping/feedback) */
/* ADCCONTROL2 (0x0A): input select. The K50835F SLIC handset transmit audio is
* wired to LIN2/RIN2 on this bench — PROVEN: speech captured (ACrms 196, crest 10.3)
* on 0x50, vs DC-only floating offset on 0x00 (LIN1). */
@@ -224,22 +224,40 @@ esp_err_t es8388_init(void)
es8388_read_reg(ES8388_CHIP_POWER, &chippower);
ESP_LOGI(TAG, "ES8388 regs: CTL1=0x%02X ADCPWR=0x%02X ADCINSEL=0x%02X ADCCTL3=0x%02X DACCTL21=0x%02X CHIPPOWER=0x%02X",
ctl1, adcpwr, adcinsel, adcctl3, dacctl21, chippower);
ESP_LOGI(TAG, "ES8388 init OK — PA enabled, DAC @ 0dB, ADC PGA +24dB, input=LIN1/RIN1 (LINE IN), DACCTL21=0x80");
ESP_LOGI(TAG, "ES8388 init OK — PA enabled, DAC @ 0dB, ADC PGA +12dB, input=LIN2/RIN2 (SLIC handset), DACCTL21=0x80");
return ESP_OK;
}
esp_err_t es8388_set_volume(uint8_t vol)
{
/* Map 0..100 to 0x00 (0dB) .. 0x24 (mute); register is attenuation.
* Volume registers: DACCONTROL24 (0x2E) = OUT1L, DACCONTROL25 (0x2F) = OUT1R,
* DACCONTROL26 (0x30) = OUT2L, DACCONTROL27 (0x31) = OUT2R.
/* ES8388 OUTx volume registers are GAIN, not attenuation: 0x00 = -45 dB
* (min) .. 0x21 = 0 dB (max); higher value = louder (>0x21 = mute/reserved).
* Map 0..100 → 0x00..0x21. (The previous code inverted this, so vol=100
* produced 0x00 = quietest — confirmed at the bench.)
* DACCONTROL24 (0x2E)=OUT1L, 25 (0x2F)=OUT1R, 26 (0x30)=OUT2L, 27 (0x31)=OUT2R.
* DACCONTROL21 (0x2B) is the ADC/DAC LRCK sync register — DO NOT touch here. */
uint8_t reg_val = (uint8_t)((100 - (int)vol) * 0x24 / 100);
ESP_LOGI(TAG, "set_volume: %d%% -> reg=0x%02X", vol, reg_val);
if (vol > 100) vol = 100;
uint8_t reg_val = (uint8_t)((int)vol * 0x21 / 100);
if (reg_val > 0x21) reg_val = 0x21;
ESP_LOGI(TAG, "set_volume: %d%% -> reg=0x%02X (0x21=max,0dB)", vol, reg_val);
esp_err_t r = ESP_OK;
r |= i2c_write_reg(ES8388_DAC_CTL24, reg_val); /* OUT1 L volume */
r |= i2c_write_reg(ES8388_DAC_CTL25, reg_val); /* OUT1 R volume */
r |= i2c_write_reg(ES8388_DAC_CTL26, reg_val); /* OUT2 L volume */
r |= i2c_write_reg(0x31, reg_val); /* OUT2 R volume (DACCONTROL27) */
return r;
}
esp_err_t es8388_set_dac_volume(uint8_t atten)
{
/* DACCONTROL4 (0x04) = LDACVOL, DACCONTROL5 (0x05) = RDACVOL: DIGITAL DAC
* volume, applied BEFORE the analog output stages. 0x00 = 0 dB, each step
* = -0.5 dB, up to 0xC0 = -96 dB (mute). Lowering this gives analog
* headroom while keeping the output-stage volume (es8388_set_volume) high. */
if (atten > 0xC0) atten = 0xC0;
ESP_LOGI(TAG, "set_dac_volume: atten=0x%02X (-%.1f dB)", atten, atten * 0.5f);
esp_err_t r = i2c_write_reg(ES8388_DAC_CTL4, atten);
r |= i2c_write_reg(ES8388_DAC_CTL5, atten);
return r;
}
+5
View File
@@ -29,6 +29,11 @@ esp_err_t es8388_init(void);
* OUT1 (headphone) + OUT2 (speaker) attenuation registers. */
esp_err_t es8388_set_volume(uint8_t vol);
/* Set the DIGITAL DAC volume (DACCONTROL4/5), applied before the analog output
* stages. atten: 0 = 0 dB (full), each step -0.5 dB, 0xC0 = mute. Lowering it
* gives analog headroom while keeping es8388_set_volume() high. */
esp_err_t es8388_set_dac_volume(uint8_t atten);
/* Mute / unmute DAC output. */
esp_err_t es8388_mute(bool mute);
+174 -3
View File
@@ -21,7 +21,6 @@
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <sys/stat.h>
#include <errno.h>
#include "audio.h"
@@ -37,9 +36,9 @@
#include "esp_log.h"
#include "esp_netif.h"
#include "esp_spiffs.h"
#include "esp_timer.h"
#include "esp_wifi.h"
#include "freertos/FreeRTOS.h"
#include "freertos/event_groups.h"
#include "freertos/task.h"
#include "nvs.h"
#include "nvs_flash.h"
@@ -467,6 +466,153 @@ static esp_err_t handle_debug_slic(httpd_req_t *req)
return send_json(req, "200 OK", buf);
}
/* ── GET /debug/hookmon (fast-sample the hook GPIO, log transitions) ─────────
* Samples CONFIG_PLIP_HOOK_GPIO every 2 ms for ~6 s and records every level
* transition with a timestamp. Catches both slow hook toggles AND fast rotary
* pulses (~60-100 ms) that the 0.7 s HTTP poll of /debug/slic cannot see.
* The user performs the physical action (lift/hang/dial) during the window. */
#define HOOKMON_SAMPLE_MS 2
#define HOOKMON_WINDOW_MS 6000
#define HOOKMON_MAX_EDGES 64
static esp_err_t handle_debug_hookmon(httpd_req_t *req)
{
const int gpio = CONFIG_PLIP_HOOK_GPIO;
int last = gpio_get_level(gpio);
const int initial = last;
int64_t t0 = esp_timer_get_time();
int edges_t[HOOKMON_MAX_EDGES];
int edges_l[HOOKMON_MAX_EDGES];
int n_edges = 0;
int lo = last, hi = last; /* track min/max seen */
ESP_LOGI(TAG, "hookmon: start, gpio=%d initial=%d (sample %dms / window %dms)",
gpio, initial, HOOKMON_SAMPLE_MS, HOOKMON_WINDOW_MS);
for (;;) {
int64_t now = esp_timer_get_time();
int dt = (int)((now - t0) / 1000);
if (dt >= HOOKMON_WINDOW_MS) break;
int lvl = gpio_get_level(gpio);
if (lvl < lo) lo = lvl;
if (lvl > hi) hi = lvl;
if (lvl != last) {
if (n_edges < HOOKMON_MAX_EDGES) {
edges_t[n_edges] = dt;
edges_l[n_edges] = lvl;
n_edges++;
}
last = lvl;
}
vTaskDelay(pdMS_TO_TICKS(HOOKMON_SAMPLE_MS));
}
/* Build JSON: { gpio, initial, final, lo, hi, edges:[{t,l},...], count } */
char buf[1024];
int off = 0;
off += snprintf(buf + off, sizeof(buf) - off,
"{\"gpio\":%d,\"initial\":%d,\"final\":%d,\"lo\":%d,\"hi\":%d,"
"\"count\":%d,\"edges\":[",
gpio, initial, last, lo, hi, n_edges);
for (int i = 0; i < n_edges && off < (int)sizeof(buf) - 32; i++) {
off += snprintf(buf + off, sizeof(buf) - off, "%s{\"t\":%d,\"l\":%d}",
i ? "," : "", edges_t[i], edges_l[i]);
}
off += snprintf(buf + off, sizeof(buf) - off, "]}");
ESP_LOGI(TAG, "hookmon: done, %d edges, lo=%d hi=%d", n_edges, lo, hi);
return send_json(req, "200 OK", buf);
}
/* ── GET /debug/dacvol?a=N (set ES8388 DIGITAL DAC volume, live tuning) ─────── */
static esp_err_t handle_debug_dacvol(httpd_req_t *req)
{
char query[32] = {0};
httpd_req_get_url_query_str(req, query, sizeof(query));
char astr[8] = {0};
if (httpd_query_key_value(query, "a", astr, sizeof(astr)) != ESP_OK) {
return send_json(req, "400 Bad Request", "{\"error\":\"missing a param (atten steps, 0=0dB)\"}");
}
int a = atoi(astr);
if (a < 0) a = 0;
if (a > 192) a = 192;
esp_err_t r = es8388_set_dac_volume((uint8_t)a);
char resp[80];
snprintf(resp, sizeof(resp), "{\"ok\":%s,\"atten_steps\":%d,\"db\":-%.1f}",
(r == ESP_OK) ? "true" : "false", a, a * 0.5);
ESP_LOGI(TAG, "debug/dacvol: atten=%d (-%.1f dB)", a, a * 0.5);
return send_json(req, "200 OK", resp);
}
/* ── GET /debug/offhook?on=1 (force hook state, bypass flaky contact) ───────── */
static esp_err_t handle_debug_offhook(httpd_req_t *req)
{
char query[24] = {0};
httpd_req_get_url_query_str(req, query, sizeof(query));
char on[4] = {0};
int v = 1; /* default: force off-hook */
if (httpd_query_key_value(query, "on", on, sizeof(on)) == ESP_OK) v = atoi(on);
phone_force_offhook(v != 0);
char resp[64];
snprintf(resp, sizeof(resp), "{\"ok\":true,\"forced_offhook\":%s}", v ? "true" : "false");
ESP_LOGI(TAG, "debug/offhook: forced %s", v ? "off-hook" : "on-hook");
return send_json(req, "200 OK", resp);
}
/* ── GET /debug/getfile?path=/x.wav (read a SPIFFS file back for diagnosis) ── */
static esp_err_t handle_debug_getfile(httpd_req_t *req)
{
char query[160] = {0};
httpd_req_get_url_query_str(req, query, sizeof(query));
char fpath[96] = {0};
if (httpd_query_key_value(query, "path", fpath, sizeof(fpath)) != ESP_OK) {
return send_json(req, "400 Bad Request", "{\"error\":\"missing path param\"}");
}
ensure_spiffs();
char full[160];
if (fpath[0] == '/') snprintf(full, sizeof(full), "%s%s", SPIFFS_BASE, fpath);
else snprintf(full, sizeof(full), "%s/%s", SPIFFS_BASE, fpath);
FILE *f = fopen(full, "rb");
if (!f) return send_json(req, "404 Not Found", "{\"error\":\"not found\"}");
httpd_resp_set_type(req, "application/octet-stream");
char chunk[1024];
size_t r;
while ((r = fread(chunk, 1, sizeof(chunk), f)) > 0) {
if (httpd_resp_send_chunk(req, chunk, r) != ESP_OK) { fclose(f); return ESP_FAIL; }
}
fclose(f);
httpd_resp_send_chunk(req, NULL, 0); /* terminate */
return ESP_OK;
}
/* ── GET /debug/vol?v=N (set ES8388 output volume 0..100, live tuning) ─────── */
static esp_err_t handle_debug_vol(httpd_req_t *req)
{
char query[32] = {0};
httpd_req_get_url_query_str(req, query, sizeof(query));
char vstr[8] = {0};
if (httpd_query_key_value(query, "v", vstr, sizeof(vstr)) != ESP_OK) {
return send_json(req, "400 Bad Request", "{\"error\":\"missing v param (0..100)\"}");
}
int v = atoi(vstr);
if (v < 0) v = 0;
if (v > 100) v = 100;
esp_err_t r = es8388_set_volume((uint8_t)v);
char resp[64];
snprintf(resp, sizeof(resp), "{\"ok\":%s,\"volume\":%d}", (r == ESP_OK) ? "true" : "false", v);
ESP_LOGI(TAG, "debug/vol: set volume=%d (%s)", v, (r == ESP_OK) ? "ok" : "err");
return send_json(req, "200 OK", resp);
}
/* ── GET /debug/dial?number=NNNN (push digits into the dialer) ───────────── */
static esp_err_t handle_debug_dial(httpd_req_t *req)
@@ -565,7 +711,7 @@ esp_err_t net_init(void)
/* Start HTTP server. */
httpd_config_t hcfg = HTTPD_DEFAULT_CONFIG();
hcfg.server_port = 80;
hcfg.max_uri_handlers = 17;
hcfg.max_uri_handlers = 16;
hcfg.stack_size = 8192;
esp_err_t ret = httpd_start(&s_httpd, &hcfg);
@@ -602,6 +748,22 @@ esp_err_t net_init(void)
.uri = "/debug/dial", .method = HTTP_GET,
.handler = handle_debug_dial, .user_ctx = NULL,
};
static const httpd_uri_t uri_debug_vol = {
.uri = "/debug/vol", .method = HTTP_GET,
.handler = handle_debug_vol, .user_ctx = NULL,
};
static const httpd_uri_t uri_debug_getfile = {
.uri = "/debug/getfile", .method = HTTP_GET,
.handler = handle_debug_getfile, .user_ctx = NULL,
};
static const httpd_uri_t uri_debug_dacvol = {
.uri = "/debug/dacvol", .method = HTTP_GET,
.handler = handle_debug_dacvol, .user_ctx = NULL,
};
static const httpd_uri_t uri_debug_offhook = {
.uri = "/debug/offhook", .method = HTTP_GET,
.handler = handle_debug_offhook, .user_ctx = NULL,
};
static const httpd_uri_t uri_debug_ring = {
.uri = "/debug/ring", .method = HTTP_GET,
.handler = handle_debug_ring, .user_ctx = NULL,
@@ -614,6 +776,10 @@ esp_err_t net_init(void)
.uri = "/debug/slic", .method = HTTP_GET,
.handler = handle_debug_slic, .user_ctx = NULL,
};
static const httpd_uri_t uri_debug_hookmon = {
.uri = "/debug/hookmon", .method = HTTP_GET,
.handler = handle_debug_hookmon, .user_ctx = NULL,
};
httpd_register_uri_handler(s_httpd, &uri_status);
httpd_register_uri_handler(s_httpd, &uri_scenario);
httpd_register_uri_handler(s_httpd, &uri_file);
@@ -621,9 +787,14 @@ esp_err_t net_init(void)
httpd_register_uri_handler(s_httpd, &uri_capture);
httpd_register_uri_handler(s_httpd, &uri_debug_regs);
httpd_register_uri_handler(s_httpd, &uri_debug_dial);
httpd_register_uri_handler(s_httpd, &uri_debug_vol);
httpd_register_uri_handler(s_httpd, &uri_debug_getfile);
httpd_register_uri_handler(s_httpd, &uri_debug_dacvol);
httpd_register_uri_handler(s_httpd, &uri_debug_offhook);
httpd_register_uri_handler(s_httpd, &uri_debug_ring);
httpd_register_uri_handler(s_httpd, &uri_debug_ringstop);
httpd_register_uri_handler(s_httpd, &uri_debug_slic);
httpd_register_uri_handler(s_httpd, &uri_debug_hookmon);
ESP_LOGI(TAG, "httpd up on :80 (GET /status, GET /debug/regs, GET /debug/dial, GET /debug/ring, GET /debug/ringstop, GET /debug/slic, POST /game/scenario, POST /game/file, POST /game/cmd, POST /voice/capture)");
return ESP_OK;
+117 -15
View File
@@ -34,13 +34,35 @@
#define TAG "phone"
#define DEBOUNCE_MS 30
#define HANGUP_THRESHOLD_MS 500 /* open > this → real hangup, not a pulse */
#define HANGUP_THRESHOLD_MS 1500 /* open > this → real hangup, not a pulse.
* Raised from 500 ms: the A1S cradle contact
* is marginal and produces ~500 ms spurious
* opens that were resetting the call before
* the NPC greeting. A real hangup holds the
* line open indefinitely, so it still fires
* (~1.5 s later). Rotary pulses (~60 ms) and
* closed inter-digit gaps are unaffected. */
#define TASK_STACK 4096
#define TASK_PRIO 5
#define RESYNC_STABLE_MS 600 /* if the raw hook level stably disagrees with
* s_offhook for this long (no pulse), the edge
* detection missed/flapped a transition →
* self-correct s_offhook to the physical level */
/* Rotary pulse hardening (CONFIG_PLIP_DIAL_PULSE) */
#define PULSE_MIN_WIDTH_MS 20 /* ignore opens shorter than this (glitch filter) */
/* real rotary pulses are ≥ 40 ms open */
#define HANGUP_VERIFY_COUNT 5 /* re-read GPIO this many times before declaring */
/* hangup (guards against marginal cradle contact) */
#define HANGUP_VERIFY_STEP_MS 20 /* delay between verification reads (ms); */
/* total span = COUNT × STEP = 100 ms */
static volatile bool s_edge_pending = false;
static volatile bool s_ringing = false;
static volatile bool s_offhook = false; /* true = handset picked up */
static volatile bool s_hook_override = false; /* when true, ignore the physical hook
* and hold s_offhook at the forced value
* (debug: decouple from a flaky contact) */
/* IRAM_ATTR: ISR must live in IRAM on original ESP32. */
static void IRAM_ATTR on_hook_isr(void *arg)
@@ -114,6 +136,7 @@ static void phone_task(void *arg)
/* Read and report initial level so master state machine is in sync. */
int last_level = gpio_get_level(hook_gpio);
s_offhook = (last_level == HOOK_OFFHOOK_LEVEL);
int resync_ms = 0; /* accumulates while raw level stably disagrees with s_offhook */
ESP_LOGI(TAG, "phone task ready, hook GPIO=%d level=%d active_%s (%s)",
hook_gpio, last_level,
(HOOK_OFFHOOK_LEVEL == 1) ? "high" : "low",
@@ -140,6 +163,15 @@ static void phone_task(void *arg)
#endif
for (;;) {
/* Debug override: hold s_offhook at the forced value, ignore the
* physical hook entirely (decouples the voice loop from a flaky
* cradle contact). Set via phone_force_offhook() / GET /debug/offhook. */
if (s_hook_override) {
s_edge_pending = false;
resync_ms = 0;
vTaskDelay(pdMS_TO_TICKS(20));
continue;
}
if (s_edge_pending) {
s_edge_pending = false;
@@ -163,8 +195,14 @@ static void phone_task(void *arg)
last_close_us = esp_timer_get_time();
int64_t open_dur_ms = (last_close_us - pulse_open_us) / 1000;
if (in_pulse && open_dur_ms < HANGUP_THRESHOLD_MS) {
/* Short open: count as a rotary pulse */
if (in_pulse && open_dur_ms < PULSE_MIN_WIDTH_MS) {
/* Glitch filter: open was too brief to be a real pulse.
* Ignore — do not count as pulse and do not declare hangup. */
in_pulse = false;
ESP_LOGD(TAG, "glitch ignored (%"PRId64"ms < %dms)",
open_dur_ms, PULSE_MIN_WIDTH_MS);
} else if (in_pulse && open_dur_ms < HANGUP_THRESHOLD_MS) {
/* Valid rotary pulse duration: count it */
pulse_count++;
in_pulse = false;
ESP_LOGD(TAG, "pulse %d (open %"PRId64"ms)", pulse_count, open_dur_ms);
@@ -231,13 +269,37 @@ static void phone_task(void *arg)
}
/* Also detect prolonged open (hangup) even if no more edges arrive.
* Active-HIGH: hangup = level stays at HOOK_PULSE_OPEN (LOW) > 500 ms. */
* Active-HIGH: hangup = level stays at HOOK_PULSE_OPEN (LOW) > 500 ms.
*
* Hardening: re-sample the GPIO HANGUP_VERIFY_COUNT times with
* HANGUP_VERIFY_STEP_MS gaps. Only declare hangup if every sample
* confirms the open state — guards against a marginal cradle contact
* that briefly dips below HANGUP_THRESHOLD_MS and then bounces back. */
if (s_offhook && in_pulse && pulse_open_us > 0) {
int64_t open_ms = (esp_timer_get_time() - pulse_open_us) / 1000;
if (open_ms >= HANGUP_THRESHOLD_MS) {
int level_now = gpio_get_level(hook_gpio);
if (level_now == HOOK_PULSE_OPEN) {
ESP_LOGI(TAG, "on-hook (prolonged open %"PRId64"ms) detected", open_ms);
/* Multi-sample verification */
bool confirmed = true;
for (int v = 0; v < HANGUP_VERIFY_COUNT; v++) {
vTaskDelay(pdMS_TO_TICKS(HANGUP_VERIFY_STEP_MS));
if (gpio_get_level(hook_gpio) != HOOK_PULSE_OPEN) {
/* Line recovered during verification — not a real hangup */
confirmed = false;
ESP_LOGD(TAG, "hangup verify failed at sample %d — bounce", v);
/* Treat the level as if the line just closed */
last_level = HOOK_PULSE_CLOSED;
last_close_us = esp_timer_get_time();
/* The open was shorter than HANGUP_THRESHOLD_MS in effect,
* but we already passed the threshold — do not count as
* a pulse (it's the same ambiguous situation we reject
* elsewhere). Simply clear in_pulse. */
in_pulse = false;
break;
}
}
if (confirmed) {
ESP_LOGI(TAG, "on-hook (prolonged open %"PRId64"ms, verified ×%d) detected",
open_ms, HANGUP_VERIFY_COUNT);
last_level = HOOK_PULSE_OPEN;
pulse_count = 0;
in_pulse = false;
@@ -245,20 +307,41 @@ static void phone_task(void *arg)
audio_stop();
audio_pa_set(false);
report_offhook(false);
} else {
/* GPIO returned to off-hook level but ISR missed it */
last_level = HOOK_PULSE_CLOSED;
last_close_us = esp_timer_get_time();
if ((esp_timer_get_time() - pulse_open_us) / 1000 < HANGUP_THRESHOLD_MS) {
pulse_count++;
}
in_pulse = false;
}
}
}
poll_sleep:
#endif
/* ── Self-healing resync ──────────────────────────────────────────────
* The marginal A1S cradle contact occasionally makes the edge detection
* miss/flap a transition, leaving s_offhook stuck (e.g. firmware thinks
* on-hook while the handset is physically off-hook → the call never
* starts). If the raw level stably disagrees with s_offhook for
* RESYNC_STABLE_MS — and we are NOT mid rotary-pulse — trust the
* physical level and correct s_offhook, firing the transition. */
{
bool pulse_active = false;
#if CONFIG_PLIP_DIAL_PULSE
pulse_active = in_pulse;
#endif
bool phys_off = (gpio_get_level(hook_gpio) == HOOK_OFFHOOK_LEVEL);
if (!pulse_active && phys_off != s_offhook) {
resync_ms += 10;
if (resync_ms >= RESYNC_STABLE_MS) {
ESP_LOGW(TAG, "hook resync: physical=%s but s_offhook=%d — correcting",
phys_off ? "off-hook" : "on-hook", (int)s_offhook);
s_offhook = phys_off;
last_level = phys_off ? HOOK_OFFHOOK_LEVEL : !HOOK_OFFHOOK_LEVEL;
audio_pa_set(phys_off);
if (!phys_off) audio_stop();
report_offhook(phys_off);
resync_ms = 0;
}
} else {
resync_ms = 0;
}
}
vTaskDelay(pdMS_TO_TICKS(10));
}
}
@@ -275,3 +358,22 @@ bool phone_is_offhook(void)
{
return s_offhook;
}
void phone_force_offhook(bool off)
{
/* Debug: override the physical hook. off=true → simulate pickup (DIALTONE),
* off=false → simulate hangup (IDLE). Stays in effect until reboot. */
s_hook_override = true;
if (off == s_offhook) {
ESP_LOGI(TAG, "force_offhook: already %s (override on)", off ? "off-hook" : "on-hook");
return;
}
s_offhook = off;
ESP_LOGI(TAG, "force_offhook: %s (override)", off ? "off-hook" : "on-hook");
audio_pa_set(off);
if (!off) {
audio_stop();
phone_ring_stop();
}
report_offhook(off);
}
+5
View File
@@ -31,6 +31,11 @@ void phone_ring_stop(void);
* Safe to call from any task; backed by a volatile flag updated in phone_task. */
bool phone_is_offhook(void);
/* Debug: force the hook state, overriding the physical contact until reboot.
* off=true simulates pickup (→ DIALTONE), off=false simulates hangup (→ IDLE).
* Lets the voice loop be validated independently of a flaky cradle contact. */
void phone_force_offhook(bool off);
#ifdef __cplusplus
}
#endif
+59 -18
View File
@@ -20,33 +20,68 @@
#define TAG "slic"
/* K50835F SHK is OPEN-COLLECTOR ACTIVE-LOW: the SLIC pulls SHK LOW when the
* loop is closed (off-hook) and releases it (pull-up → HIGH) when on-hook.
* So off-hook = LOW. (A252 used active-high via an external inverter; the A1S
* wires SHK straight to the GPIO, so the raw polarity is active-low.) */
#define SLIC_SHK_OFFHOOK_LEVEL 0
/* Empirically on THIS A1S+SLIC unit, SHK is ACTIVE-HIGH: off-hook (loop closed)
* drives SHK HIGH, on-hook drives it LOW. Confirmed at the bench via
* /debug/hookmon: on-hook→0, pickup-during-ring→0→1 transition, off-hook→1.
* (The A252 reference was active-low; the polarity is inverted here, so the
* raw straight-wired GPIO reads off-hook = HIGH.) */
#define SLIC_SHK_OFFHOOK_LEVEL 1
static volatile bool s_ringing = false;
static TaskHandle_t s_ring_task = NULL;
/* ── FR toggle task (ring drive at ~25 Hz) ────────────────────────────────── */
/* France Télécom ring cadence: 1.5 s burst ON, 3.5 s silent pause, repeating.
* (Single-ring cadence — distinct from the UK double-ring or US 2 s/4 s.)
* The bell is driven during the burst by toggling FR at ~25 Hz with RM HIGH;
* during the pause RM and FR are held low so the bell is silent. */
#define RING_FR_TOGGLE_MS 20 /* 20 ms half-period → ~25 Hz bell drive */
#define RING_BURST_MS 1500 /* FT: 1.5 s of ringing */
#define RING_PAUSE_MS 3500 /* FT: 3.5 s of silence */
/* ── FR toggle + cadence task ─────────────────────────────────────────────── */
static void slic_ring_task(void *arg)
{
(void)arg;
bool fr_state = false;
bool in_burst = true; /* start each ring with a burst */
int phase_ms = 0; /* elapsed ms in the current burst/pause phase */
for (;;) {
if (s_ringing) {
fr_state = !fr_state;
gpio_set_level(PLIP_SLIC_FR, fr_state ? 1 : 0);
ESP_LOGV(TAG, "FR=%d", fr_state ? 1 : 0);
if (in_burst) {
/* Ring burst: RM HIGH, FR toggling at ~25 Hz to swing the bell */
gpio_set_level(PLIP_SLIC_RM, 1);
fr_state = !fr_state;
gpio_set_level(PLIP_SLIC_FR, fr_state ? 1 : 0);
phase_ms += RING_FR_TOGGLE_MS;
if (phase_ms >= RING_BURST_MS) {
/* End of burst → enter silent pause */
in_burst = false;
phase_ms = 0;
fr_state = false;
gpio_set_level(PLIP_SLIC_RM, 0);
gpio_set_level(PLIP_SLIC_FR, 0);
ESP_LOGV(TAG, "ring: burst end -> pause");
}
} else {
/* Silent pause: RM/FR stay low */
phase_ms += RING_FR_TOGGLE_MS;
if (phase_ms >= RING_PAUSE_MS) {
in_burst = true;
phase_ms = 0;
ESP_LOGV(TAG, "ring: pause end -> burst");
}
}
} else {
/* Suspended — keep FR low while idle */
/* Idle — keep RM/FR low and reset the cadence for the next ring */
gpio_set_level(PLIP_SLIC_RM, 0);
gpio_set_level(PLIP_SLIC_FR, 0);
fr_state = false;
in_burst = true;
phase_ms = 0;
}
vTaskDelay(pdMS_TO_TICKS(20)); /* 20 ms → ~25 Hz */
vTaskDelay(pdMS_TO_TICKS(RING_FR_TOGGLE_MS));
}
}
@@ -54,10 +89,12 @@ static void slic_ring_task(void *arg)
esp_err_t slic_init(void)
{
/* RM: Ring Mode output, init LOW */
/* RM: Ring Mode output, init LOW.
* INPUT_OUTPUT (not plain OUTPUT) so gpio_get_level() reads back the real
* pad level — plain OUTPUT disables the input buffer and always reads 0. */
gpio_config_t rm_cfg = {
.pin_bit_mask = (1ULL << PLIP_SLIC_RM),
.mode = GPIO_MODE_OUTPUT,
.mode = GPIO_MODE_INPUT_OUTPUT,
.pull_up_en = GPIO_PULLUP_DISABLE,
.pull_down_en = GPIO_PULLDOWN_DISABLE,
.intr_type = GPIO_INTR_DISABLE,
@@ -66,10 +103,10 @@ esp_err_t slic_init(void)
if (ret != ESP_OK) return ret;
gpio_set_level(PLIP_SLIC_RM, 0);
/* FR: Forward/Reverse output, init LOW */
/* FR: Forward/Reverse output, init LOW (INPUT_OUTPUT for readback, see RM). */
gpio_config_t fr_cfg = {
.pin_bit_mask = (1ULL << PLIP_SLIC_FR),
.mode = GPIO_MODE_OUTPUT,
.mode = GPIO_MODE_INPUT_OUTPUT,
.pull_up_en = GPIO_PULLUP_DISABLE,
.pull_down_en = GPIO_PULLDOWN_DISABLE,
.intr_type = GPIO_INTR_DISABLE,
@@ -92,7 +129,9 @@ esp_err_t slic_init(void)
/* PD: Power Down — EXACT A252-proven sequence: open-drain output, HIGH = released
* = SLIC active (setPowerDown(false) in Ks0835SlicController). This is the config
* the working Arduino slic-phone project uses on the same chip. */
ret = gpio_set_direction(PLIP_SLIC_PD, GPIO_MODE_OUTPUT_OD);
/* INPUT_OUTPUT_OD (not plain OUTPUT_OD) so gpio_get_level() reads the real
* pad level — OUTPUT_OD also disables the input buffer and would read 0. */
ret = gpio_set_direction(PLIP_SLIC_PD, GPIO_MODE_INPUT_OUTPUT_OD);
if (ret != ESP_OK) return ret;
ret = gpio_set_level(PLIP_SLIC_PD, 1); /* open-drain released HIGH = active (A252-proven) */
if (ret != ESP_OK) return ret;
@@ -123,8 +162,10 @@ bool slic_is_offhook(void)
void slic_ring_start(void)
{
if (s_ringing) return;
ESP_LOGI(TAG, "ring start: RM=HIGH, FR toggling at 25 Hz");
gpio_set_level(PLIP_SLIC_RM, 1);
ESP_LOGI(TAG, "ring start: France Télécom cadence %d ms ON / %d ms OFF",
RING_BURST_MS, RING_PAUSE_MS);
/* The cadence task drives RM/FR; it starts on a burst (in_burst reset in
* the idle branch). Just arm it here. */
s_ringing = true;
}
+211 -2
View File
@@ -1,11 +1,14 @@
/*
* turn_client.c — NPC gateway client for /v1/voice/turn.
* turn_client.c — NPC gateway client for /v1/voice/turn and /v1/voice/reply.
*
* Uses esp_http_client open/write/fetch/read streaming so the binary WAV
* body can be written directly to SPIFFS without a large heap buffer.
*
* Pattern mirrors hook_client.c but synchronous (called from conv_task).
* Timeout is generous (30 s) because TTS synthesis adds latency.
*
* Stage 3 adds turn_client_reply() which POSTs captured mic audio as
* multipart/form-data to /v1/voice/reply and streams the NPC response WAV.
*/
#include "turn_client.h"
@@ -26,8 +29,17 @@
/* Read chunks when streaming the response body. */
#define CHUNK_SIZE 1024
/* Write chunks when uploading the captured WAV body (4 KB). */
#define UPLOAD_CHUNK 4096
/* Timeout for the full TTS round-trip (connect + synthesis + transfer). */
#define TIMEOUT_MS 30000
#define TIMEOUT_MS 90000 /* reply TTS (Kyutai MLX ~0.3x realtime) can take
* tens of seconds; was 30s → reply POST timed out
* (HTTP -1). 90s covers worst-case generation. */
/* Multipart boundary (must not appear in the WAV payload — 16kHz PCM is binary
* so any fixed ASCII boundary is safe). */
#define BOUNDARY "----ZacusPlipBoundary7MA4YWxkTrZu0gW"
bool turn_client_greeting(const char *session_id,
const char *number,
@@ -129,3 +141,200 @@ bool turn_client_greeting(const char *session_id,
ESP_LOGI(TAG, "greeting WAV %d bytes written to %s", total, out_path);
return true;
}
/* ── turn_client_reply ─────────────────────────────────────────────────────
*
* POST multipart/form-data to /v1/voice/reply with the captured mic WAV.
*
* Multipart layout (each part is CRLF-delimited per RFC 2046):
*
* --<boundary>\r\n
* Content-Disposition: form-data; name="session_id"\r\n\r\n
* <session_id>\r\n
* --<boundary>\r\n
* Content-Disposition: form-data; name="number"\r\n\r\n
* <number>\r\n
* --<boundary>\r\n
* Content-Disposition: form-data; name="audio"; filename="rec.wav"\r\n
* Content-Type: audio/wav\r\n\r\n
* <wav bytes>
* \r\n--<boundary>--\r\n
*
* Content-Length = len(preamble) + wav_len + len(epilogue)
* (known exactly before writing — allows non-chunked POST).
*/
esp_err_t turn_client_reply(const char *session_id,
const char *number,
const uint8_t *wav,
size_t wav_len,
const char *out_path)
{
if (!session_id || !number || !wav || wav_len == 0 || !out_path) {
return ESP_ERR_INVALID_ARG;
}
/* --- Build URL -------------------------------------------------------- */
char url[256];
snprintf(url, sizeof(url), "%s/v1/voice/reply",
CONFIG_PLIP_GATEWAY_URL);
/* --- Build multipart preamble ---------------------------------------- */
/* preamble = 3 parts before the raw wav bytes:
* part 1 — session_id (text field)
* part 2 — number (text field)
* part 3 — audio file header (up to but NOT including the file body)
*/
char preamble[512];
int preamble_len = snprintf(preamble, sizeof(preamble),
"--%s\r\n"
"Content-Disposition: form-data; name=\"session_id\"\r\n\r\n"
"%s\r\n"
"--%s\r\n"
"Content-Disposition: form-data; name=\"number\"\r\n\r\n"
"%s\r\n"
"--%s\r\n"
"Content-Disposition: form-data; name=\"audio\"; filename=\"rec.wav\"\r\n"
"Content-Type: audio/wav\r\n\r\n",
BOUNDARY,
session_id,
BOUNDARY,
number,
BOUNDARY);
if (preamble_len <= 0 || preamble_len >= (int)sizeof(preamble)) {
ESP_LOGE(TAG, "preamble buffer overflow (len=%d)", preamble_len);
return ESP_ERR_INVALID_SIZE;
}
/* --- Build multipart epilogue ---------------------------------------- */
/* epilogue = CRLF after the wav data + closing boundary */
char epilogue[64];
int epilogue_len = snprintf(epilogue, sizeof(epilogue),
"\r\n--%s--\r\n",
BOUNDARY);
if (epilogue_len <= 0 || epilogue_len >= (int)sizeof(epilogue)) {
ESP_LOGE(TAG, "epilogue buffer overflow");
return ESP_ERR_INVALID_SIZE;
}
int content_length = preamble_len + (int)wav_len + epilogue_len;
ESP_LOGI(TAG, "reply POST %s preamble=%d wav=%zu epilogue=%d total=%d",
url, preamble_len, wav_len, epilogue_len, content_length);
/* --- Configure client ------------------------------------------------- */
esp_http_client_config_t cfg = {
.url = url,
.method = HTTP_METHOD_POST,
.timeout_ms = TIMEOUT_MS,
};
esp_http_client_handle_t client = esp_http_client_init(&cfg);
if (!client) {
ESP_LOGE(TAG, "esp_http_client_init failed");
return ESP_ERR_NO_MEM;
}
/* Content-Type with boundary */
char ct_header[128];
snprintf(ct_header, sizeof(ct_header),
"multipart/form-data; boundary=%s", BOUNDARY);
esp_http_client_set_header(client, "Content-Type", ct_header);
/* Authorization header — skip if token is empty */
const char *token = CONFIG_PLIP_GATEWAY_TOKEN;
if (token && token[0] != '\0') {
char auth[128];
snprintf(auth, sizeof(auth), "Bearer %s", token);
esp_http_client_set_header(client, "Authorization", auth);
}
/* --- Open connection and stream body ---------------------------------- */
esp_err_t err = esp_http_client_open(client, content_length);
if (err != ESP_OK) {
ESP_LOGW(TAG, "open %s failed: %s", url, esp_err_to_name(err));
esp_http_client_cleanup(client);
return err;
}
/* Write preamble */
int written = esp_http_client_write(client, preamble, preamble_len);
if (written < 0) {
ESP_LOGW(TAG, "write preamble failed (ret=%d)", written);
esp_http_client_close(client);
esp_http_client_cleanup(client);
return ESP_FAIL;
}
/* Write WAV data in 4 KB chunks */
size_t remaining = wav_len;
const uint8_t *ptr = wav;
while (remaining > 0) {
size_t to_send = (remaining < UPLOAD_CHUNK) ? remaining : UPLOAD_CHUNK;
int wr = esp_http_client_write(client, (const char *)ptr, (int)to_send);
if (wr < 0) {
ESP_LOGW(TAG, "write wav chunk failed (ret=%d)", wr);
esp_http_client_close(client);
esp_http_client_cleanup(client);
return ESP_FAIL;
}
ptr += (size_t)wr;
remaining -= (size_t)wr;
}
/* Write epilogue */
written = esp_http_client_write(client, epilogue, epilogue_len);
if (written < 0) {
ESP_LOGW(TAG, "write epilogue failed (ret=%d)", written);
esp_http_client_close(client);
esp_http_client_cleanup(client);
return ESP_FAIL;
}
/* --- Fetch response headers ------------------------------------------ */
int content_len = esp_http_client_fetch_headers(client);
int status_code = esp_http_client_get_status_code(client);
ESP_LOGI(TAG, "reply HTTP %d content_length=%d", status_code, content_len);
if (status_code != 200) {
ESP_LOGW(TAG, "gateway returned HTTP %d for reply", status_code);
esp_http_client_close(client);
esp_http_client_cleanup(client);
return ESP_ERR_INVALID_RESPONSE;
}
/* --- Stream binary WAV response into SPIFFS file --------------------- */
FILE *fp = fopen(out_path, "wb");
if (!fp) {
ESP_LOGE(TAG, "fopen(%s, wb) failed", out_path);
esp_http_client_close(client);
esp_http_client_cleanup(client);
return ESP_ERR_NOT_FOUND;
}
static uint8_t s_reply_chunk[CHUNK_SIZE]; /* static: avoids stack pressure */
int total = 0;
int rd;
while ((rd = esp_http_client_read(client, (char *)s_reply_chunk,
sizeof(s_reply_chunk))) > 0) {
size_t fw = fwrite(s_reply_chunk, 1, (size_t)rd, fp);
if ((int)fw != rd) {
ESP_LOGW(TAG, "fwrite short: wrote %d of %d bytes", (int)fw, rd);
break;
}
total += rd;
}
fclose(fp);
esp_http_client_close(client);
esp_http_client_cleanup(client);
if (total <= MIN_WAV_BYTES) {
ESP_LOGW(TAG, "reply WAV too short (%d bytes) — ignoring", total);
return ESP_ERR_INVALID_SIZE;
}
ESP_LOGI(TAG, "reply WAV %d bytes written to %s", total, out_path);
return ESP_OK;
}
+32 -1
View File
@@ -3,9 +3,12 @@
* turn_client.h — POST /v1/voice/turn to the NPC gateway and retrieve a WAV response.
*
* Stage 2: greeting fetch (kind="greeting").
* Stage 3 will add kind="listen" / "speak".
* Stage 3: reply fetch via /v1/voice/reply (multipart/form-data with captured WAV).
*/
#include <stdbool.h>
#include <stdint.h>
#include <stddef.h>
#include "esp_err.h"
#ifdef __cplusplus
extern "C" {
@@ -30,6 +33,34 @@ bool turn_client_greeting(const char *session_id,
const char *number,
const char *out_path);
/*
* turn_client_reply — POST /v1/voice/reply as multipart/form-data.
*
* Endpoint: CONFIG_PLIP_GATEWAY_URL/v1/voice/reply
* Method: POST multipart/form-data
* Fields: session_id (text), number (text), audio (file, "rec.wav", audio/wav)
* Bearer: CONFIG_PLIP_GATEWAY_TOKEN (skipped if empty)
*
* The function builds the multipart body in three segments:
* preamble = boundary + session_id part + boundary + number part +
* boundary + audio file header
* wav data = wav bytes (wav_len bytes, sent in 4 KB chunks)
* epilogue = CRLF + closing boundary
*
* Content-Length = len(preamble) + wav_len + len(epilogue)
* The response body (WAV 16 kHz) is streamed into out_path.
*
* Response headers X-Zacus-Heard and X-Zacus-Said are logged at INFO level.
*
* Returns ESP_OK if HTTP 200 and a valid WAV (> 44 bytes) was written.
* On any failure: logs a warning and returns an ESP error code.
*/
esp_err_t turn_client_reply(const char *session_id,
const char *number,
const uint8_t *wav,
size_t wav_len,
const char *out_path);
#ifdef __cplusplus
}
#endif
+1 -1
View File
@@ -37,7 +37,7 @@ CONFIG_FREERTOS_IDLE_TASK_STACKSIZE=2048
# SLIC hook GPIO — SHK on GPIO23 (A1S KEY4, active-HIGH: HIGH = off-hook)
CONFIG_PLIP_HOOK_GPIO=23
CONFIG_PLIP_HOOK_ACTIVE_HIGH=n
CONFIG_PLIP_HOOK_ACTIVE_HIGH=y
# WiFi credentials — set via `idf.py menuconfig` (PLIP Voice Configuration)
# or override locally in sdkconfig (NOT committed).