From 58a968bc9e632fc48a140afba514edf642e93fa6 Mon Sep 17 00:00:00 2001
From: cai <cai@nbcai.cc>
Date: Thu, 09 Jul 2026 23:24:45 +0800
Subject: [PATCH] feat: upload user turns to asr stream
---
src/main.rs | 536 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-----
1 files changed, 487 insertions(+), 49 deletions(-)
diff --git a/src/main.rs b/src/main.rs
index e05f602..1e969cc 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -12,7 +12,7 @@
};
use anyhow::{Context, Result, anyhow};
-use audio::{AudioDiagnostics, PcmFrame, load_pre_recorded_frames};
+use audio::{AudioDiagnostics, load_pre_recorded_frames};
use base64::{Engine as _, engine::general_purpose};
use futures_util::StreamExt;
use libwebrtc::{
@@ -34,8 +34,12 @@
use tokio::{sync::mpsc::UnboundedReceiver, task::JoinHandle};
use tracing::{info, warn};
-const TARGET_SAMPLE_RATE_HZ: u32 = 48_000;
-const TARGET_NUM_CHANNELS: u16 = 1;
+const USER_AUDIO_SAMPLE_RATE_HZ: u32 = 48_000;
+const USER_AUDIO_NUM_CHANNELS: u16 = 1;
+const DEFAULT_BOT_AUDIO_PROFILE: &str = "pcm-16k";
+const LIVEKIT_48K_SAMPLE_RATE_HZ: u32 = 48_000;
+const PCM_16K_SAMPLE_RATE_HZ: u32 = 16_000;
+const BOT_NUM_CHANNELS: u16 = 1;
const TRACK_NAME: &str = "bot-main-audio";
#[tokio::main(flavor = "multi_thread")]
@@ -58,8 +62,8 @@
&http,
config.greeting_audio_file.as_deref(),
config.greeting_audio_url.as_deref(),
- TARGET_SAMPLE_RATE_HZ,
- TARGET_NUM_CHANNELS,
+ config.bot_audio_profile.sample_rate_hz,
+ config.bot_audio_profile.num_channels,
config.audio_debug_dump_dir.as_deref(),
&config.call_id,
"greeting",
@@ -81,6 +85,9 @@
room_alias = %redact(&config.room_id),
participant_alias = %redact(&config.bot_participant_identity),
greeting_source = %config.greeting_source,
+ bot_audio_profile = %config.bot_audio_profile.profile,
+ bot_sample_rate_hz = config.bot_audio_profile.sample_rate_hz,
+ bot_num_channels = config.bot_audio_profile.num_channels,
"combrabo voice runtime helper connected"
);
emit_activity(
@@ -129,8 +136,9 @@
&config.call_id,
&config.trace_id,
TRACK_NAME,
- TARGET_SAMPLE_RATE_HZ,
- u32::from(TARGET_NUM_CHANNELS),
+ config.bot_audio_profile.profile.clone(),
+ config.bot_audio_profile.sample_rate_hz,
+ u32::from(config.bot_audio_profile.num_channels),
config.user_participant_identity.clone(),
)
.await?;
@@ -262,6 +270,8 @@
runtime_turn_bridge_url: Option<String>,
runtime_turn_bridge_token: Option<String>,
runtime_turn_bridge_mode: String,
+ runtime_asr_stream_enabled: bool,
+ runtime_asr_stream_url: Option<String>,
runtime_turn_artifact_dir: Option<String>,
runtime_session_nonce: Option<String>,
user_audio_observer_enabled: bool,
@@ -269,6 +279,54 @@
simple_vad_gate_until_greeting_done: bool,
simple_vad_post_greeting_delay_ms: u64,
simple_vad_config: SimpleVadConfig,
+ bot_audio_profile: BotAudioProfile,
+}
+
+#[derive(Clone)]
+struct BotAudioProfile {
+ profile: String,
+ sample_rate_hz: u32,
+ num_channels: u16,
+}
+
+impl BotAudioProfile {
+ fn from_env() -> Result<Self> {
+ let profile = env::var("CV_BOT_AUDIO_PROFILE")
+ .unwrap_or_else(|_| DEFAULT_BOT_AUDIO_PROFILE.to_string())
+ .trim()
+ .to_ascii_lowercase();
+ match profile.as_str() {
+ "livekit-48k" | "48k" => Ok(Self {
+ profile: "livekit-48k".to_string(),
+ sample_rate_hz: LIVEKIT_48K_SAMPLE_RATE_HZ,
+ num_channels: BOT_NUM_CHANNELS,
+ }),
+ "pcm-16k" | "16k" => Ok(Self {
+ profile: "pcm-16k".to_string(),
+ sample_rate_hz: PCM_16K_SAMPLE_RATE_HZ,
+ num_channels: BOT_NUM_CHANNELS,
+ }),
+ "custom" => {
+ let sample_rate_hz = u32_env("CV_BOT_SAMPLE_RATE_HZ", LIVEKIT_48K_SAMPLE_RATE_HZ);
+ let num_channels = u16_env("CV_BOT_NUM_CHANNELS", BOT_NUM_CHANNELS);
+ if sample_rate_hz == 0 {
+ return Err(anyhow!("CV_BOT_SAMPLE_RATE_HZ must be positive"));
+ }
+ if num_channels == 0 {
+ return Err(anyhow!("CV_BOT_NUM_CHANNELS must be positive"));
+ }
+ Ok(Self {
+ profile,
+ sample_rate_hz,
+ num_channels,
+ })
+ }
+ _ => Err(anyhow!(
+ "unsupported CV_BOT_AUDIO_PROFILE {}; expected livekit-48k, pcm-16k or custom",
+ profile
+ )),
+ }
+ }
}
#[derive(Clone)]
@@ -276,6 +334,8 @@
bridge_url: Option<String>,
bridge_token: Option<String>,
bridge_mode: String,
+ asr_stream_enabled: bool,
+ asr_stream_url: Option<String>,
artifact_dir: Option<String>,
runtime_session_nonce: Option<String>,
audio_debug_dump_dir: Option<String>,
@@ -287,6 +347,8 @@
bridge_url: config.runtime_turn_bridge_url.clone(),
bridge_token: config.runtime_turn_bridge_token.clone(),
bridge_mode: config.runtime_turn_bridge_mode.clone(),
+ asr_stream_enabled: config.runtime_asr_stream_enabled,
+ asr_stream_url: config.runtime_asr_stream_url.clone(),
artifact_dir: config.runtime_turn_artifact_dir.clone(),
runtime_session_nonce: config.runtime_session_nonce.clone(),
audio_debug_dump_dir: config.audio_debug_dump_dir.clone(),
@@ -318,6 +380,22 @@
.as_deref()
.is_some_and(|value| value.trim_end_matches('/').ends_with("/stream"))
}
+
+ fn is_asr_stream_ready(&self) -> bool {
+ self.asr_stream_enabled
+ && self
+ .asr_stream_url
+ .as_ref()
+ .is_some_and(|value| !value.is_empty())
+ && self
+ .bridge_token
+ .as_ref()
+ .is_some_and(|value| !value.is_empty())
+ && self
+ .runtime_session_nonce
+ .as_ref()
+ .is_some_and(|value| !value.is_empty())
+ }
}
impl Config {
@@ -339,6 +417,8 @@
runtime_turn_bridge_token: optional_env("CV_RUNTIME_TURN_BRIDGE_TOKEN"),
runtime_turn_bridge_mode: env::var("CV_RUNTIME_TURN_BRIDGE_MODE")
.unwrap_or_else(|_| "json".to_string()),
+ runtime_asr_stream_enabled: bool_env("CV_RUNTIME_ASR_STREAM_ENABLED", false),
+ runtime_asr_stream_url: optional_env("CV_RUNTIME_ASR_STREAM_URL"),
runtime_turn_artifact_dir: optional_env("CV_RUNTIME_TURN_ARTIFACT_DIR"),
runtime_session_nonce: optional_env("CV_RUNTIME_SESSION_NONCE"),
user_audio_observer_enabled: bool_env("CV_ENABLE_USER_AUDIO_OBSERVER", true),
@@ -346,6 +426,7 @@
simple_vad_gate_until_greeting_done: bool_env("CV_VAD_GATE_UNTIL_GREETING_DONE", true),
simple_vad_post_greeting_delay_ms: u64_env("CV_VAD_POST_GREETING_DELAY_MS", 800),
simple_vad_config: SimpleVadConfig::from_env(),
+ bot_audio_profile: BotAudioProfile::from_env()?,
})
}
}
@@ -492,6 +573,13 @@
.unwrap_or(default_value)
}
+fn u16_env(key: &str, default_value: u16) -> u16 {
+ env::var(key)
+ .ok()
+ .and_then(|value| value.trim().parse::<u16>().ok())
+ .unwrap_or(default_value)
+}
+
fn redact(value: &str) -> String {
if value.len() <= 8 {
return "redacted".to_string();
@@ -532,6 +620,7 @@
"traceId": trace_id,
"turnId": turn_id,
"eventName": event_name,
+ "eventWallTimeMs": current_time_millis(),
"result": result,
"reasonCode": reason_code,
"retryable": retryable,
@@ -742,9 +831,11 @@
"turnArtifactBytes": byte_size,
"frameCount": turn.frame_count,
"sampleCount": turn.sample_count,
- "endReason": turn.end_reason,
+ "endReason": turn.end_reason.as_str(),
}),
);
+ let asr_result_ref =
+ request_asr_result_ref(http, bridge_config, call_id, trace_id, &turn).await;
if bridge_config.is_stream_mode() {
match request_turn_bridge_stream(
http,
@@ -755,6 +846,7 @@
&turn,
&path_ref,
byte_size,
+ asr_result_ref.as_deref(),
turn_pipeline_started_at,
)
.await
@@ -818,6 +910,7 @@
&turn,
&path_ref,
byte_size,
+ asr_result_ref.as_deref(),
turn_pipeline_started_at,
)
.await
@@ -930,13 +1023,258 @@
audio::write_pcm_wav(
&output_path,
&turn.samples,
- TARGET_SAMPLE_RATE_HZ,
- TARGET_NUM_CHANNELS,
+ USER_AUDIO_SAMPLE_RATE_HZ,
+ USER_AUDIO_NUM_CHANNELS,
)?;
let byte_size = fs::metadata(&output_path)
.context("failed to stat turn artifact")?
.len();
Ok((path_ref, byte_size))
+}
+
+async fn request_asr_result_ref(
+ http: &Client,
+ bridge_config: &TurnBridgeConfig,
+ call_id: &str,
+ trace_id: &str,
+ turn: &FinishedSpeechTurn,
+) -> Option<String> {
+ if !bridge_config.is_asr_stream_ready() {
+ return None;
+ }
+ match request_asr_stream(http, bridge_config, call_id, trace_id, turn).await {
+ Ok(Some(asr_result_ref)) => {
+ emit_activity(
+ call_id,
+ trace_id,
+ Some(&turn.turn_id),
+ "asr_stream_ref_ready",
+ "ok",
+ None,
+ None,
+ json!({
+ "asrResultRefPresent": true,
+ "format": "pcm_s16le",
+ "sampleRate": 16000,
+ "channels": 1,
+ }),
+ );
+ Some(asr_result_ref)
+ }
+ Ok(None) => None,
+ Err(error) => {
+ warn!(
+ call_id = %call_id,
+ trace_id = %trace_id,
+ turn_id = %turn.turn_id,
+ error = %safe_error(&error.to_string()),
+ "runtime helper asr_stream_failed_fallback"
+ );
+ emit_activity(
+ call_id,
+ trace_id,
+ Some(&turn.turn_id),
+ "asr_stream_fallback",
+ "skipped",
+ Some("ASR_STREAM_INTERRUPTED"),
+ Some(true),
+ json!({
+ "fallbackReason": "asr_stream_request_failed",
+ "fallbackStage": "asr_stream",
+ }),
+ );
+ None
+ }
+ }
+}
+
+async fn request_asr_stream(
+ http: &Client,
+ bridge_config: &TurnBridgeConfig,
+ call_id: &str,
+ trace_id: &str,
+ turn: &FinishedSpeechTurn,
+) -> Result<Option<String>> {
+ let started_at = Instant::now();
+ let ndjson = build_asr_stream_ndjson(call_id, trace_id, turn, bridge_config)?;
+ let response = http
+ .post(bridge_config.asr_stream_url.as_deref().unwrap_or_default())
+ .header("Content-Type", "application/x-ndjson")
+ .header(
+ "X-CV-Runtime-Token",
+ bridge_config.bridge_token.as_deref().unwrap_or_default(),
+ )
+ .header("X-CV-Call-Id", call_id)
+ .header("X-CV-Trace-Id", trace_id)
+ .header(
+ "X-CV-Runtime-Session-Nonce",
+ bridge_config
+ .runtime_session_nonce
+ .as_deref()
+ .unwrap_or_default(),
+ )
+ .body(ndjson)
+ .send()
+ .await
+ .context("failed to post asr stream")?;
+ let status = response.status();
+ if !status.is_success() {
+ let body_len = response.text().await.map(|body| body.len()).unwrap_or(0);
+ warn!(
+ call_id = %call_id,
+ trace_id = %trace_id,
+ turn_id = %turn.turn_id,
+ http_status = status.as_u16(),
+ body_len,
+ "runtime helper asr_stream_http_failed"
+ );
+ return Ok(None);
+ }
+ let body: RuntimeTurnCommonResult<RuntimeAsrStreamResp> = response
+ .json()
+ .await
+ .context("failed to decode asr stream response")?;
+ if body.code != 0 {
+ warn!(
+ call_id = %call_id,
+ trace_id = %trace_id,
+ turn_id = %turn.turn_id,
+ code = body.code,
+ msg_len = body.msg.as_deref().unwrap_or_default().len(),
+ "runtime helper asr_stream_common_result_failed"
+ );
+ return Ok(None);
+ }
+ let Some(data) = body.data else {
+ return Ok(None);
+ };
+ if data.status.as_deref() == Some("final") {
+ info!(
+ call_id = %call_id,
+ trace_id = %trace_id,
+ turn_id = %turn.turn_id,
+ chunk_count = data.chunk_count.unwrap_or_default(),
+ audio_bytes = data.audio_bytes.unwrap_or_default(),
+ asr_duration_ms = data.asr_duration_ms.unwrap_or_default(),
+ wall_ms = started_at.elapsed().as_millis() as u64,
+ provider = %data.provider_alias.as_deref().unwrap_or("unknown"),
+ text_len = data.text_len.unwrap_or_default(),
+ "runtime helper asr_stream_final"
+ );
+ return Ok(data.asr_result_ref);
+ }
+ emit_activity(
+ call_id,
+ trace_id,
+ Some(&turn.turn_id),
+ "asr_stream_fallback",
+ "skipped",
+ None,
+ Some(true),
+ json!({
+ "fallbackReason": data.fallback_reason,
+ "fallbackStage": data.fallback_stage,
+ "status": data.status,
+ }),
+ );
+ Ok(None)
+}
+
+fn build_asr_stream_ndjson(
+ call_id: &str,
+ trace_id: &str,
+ turn: &FinishedSpeechTurn,
+ bridge_config: &TurnBridgeConfig,
+) -> Result<String> {
+ let chunks = asr_pcm_16k_chunks(turn)?;
+ let mut seq = 1u64;
+ let mut lines = Vec::with_capacity(chunks.len() + 3);
+ lines.push(serde_json::to_string(&json!({
+ "event": "asr_stream_started",
+ "seq": seq,
+ "callId": call_id,
+ "traceId": trace_id,
+ "turnId": turn.turn_id.as_str(),
+ "tsMs": current_time_millis(),
+ "payload": {
+ "format": "pcm_s16le",
+ "sampleRate": 16000,
+ "channels": 1,
+ "runtimeSessionNonce": bridge_config.runtime_session_nonce.as_deref().unwrap_or_default(),
+ "providerHint": "volcengine",
+ }
+ }))?);
+ for (index, samples) in chunks.iter().enumerate() {
+ seq += 1;
+ let bytes = pcm_i16_to_le_bytes(samples);
+ let duration_ms = ((samples.len() as u64) * 1000 / 16_000).max(1);
+ lines.push(serde_json::to_string(&json!({
+ "event": "asr_audio_chunk",
+ "seq": seq,
+ "callId": call_id,
+ "traceId": trace_id,
+ "turnId": turn.turn_id.as_str(),
+ "tsMs": current_time_millis(),
+ "payload": {
+ "chunkSeq": index + 1,
+ "format": "pcm_s16le",
+ "sampleRate": 16000,
+ "channels": 1,
+ "durationMs": duration_ms,
+ "payloadBase64": general_purpose::STANDARD.encode(bytes),
+ }
+ }))?);
+ }
+ seq += 1;
+ lines.push(serde_json::to_string(&json!({
+ "event": "vad_speech_end",
+ "seq": seq,
+ "callId": call_id,
+ "traceId": trace_id,
+ "turnId": turn.turn_id.as_str(),
+ "tsMs": current_time_millis(),
+ "payload": {
+ "endReason": turn.end_reason.as_str(),
+ "speechDurationMs": turn.duration_ms,
+ }
+ }))?);
+ seq += 1;
+ lines.push(serde_json::to_string(&json!({
+ "event": "asr_stream_finish",
+ "seq": seq,
+ "callId": call_id,
+ "traceId": trace_id,
+ "turnId": turn.turn_id.as_str(),
+ "tsMs": current_time_millis(),
+ "payload": {
+ "finalChunkSeq": chunks.len(),
+ "audioDurationMs": turn.duration_ms,
+ }
+ }))?);
+ Ok(lines.join("\n") + "\n")
+}
+
+fn asr_pcm_16k_chunks(turn: &FinishedSpeechTurn) -> Result<Vec<Vec<i16>>> {
+ if turn.samples.is_empty() {
+ return Err(anyhow!("empty turn samples"));
+ }
+ let samples_16k: Vec<i16> = turn.samples.iter().step_by(3).copied().collect();
+ if samples_16k.is_empty() {
+ return Err(anyhow!("empty 16k asr samples"));
+ }
+ let samples_per_chunk = 320usize;
+ Ok(samples_16k
+ .chunks(samples_per_chunk)
+ .map(|chunk| chunk.to_vec())
+ .collect())
+}
+
+fn pcm_i16_to_le_bytes(samples: &[i16]) -> Vec<u8> {
+ let mut bytes = Vec::with_capacity(samples.len() * 2);
+ for sample in samples {
+ bytes.extend_from_slice(&sample.to_le_bytes());
+ }
+ bytes
}
async fn request_turn_bridge_stream(
@@ -948,6 +1286,7 @@
turn: &FinishedSpeechTurn,
path_ref: &str,
byte_size: u64,
+ asr_result_ref: Option<&str>,
turn_pipeline_started_at: Instant,
) -> Result<RuntimeTurnStreamOutcome> {
let bridge_started_at = Instant::now();
@@ -955,15 +1294,16 @@
call_id: call_id.to_string(),
trace_id: trace_id.to_string(),
turn_id: turn.turn_id.clone(),
- audio_artifact: RuntimeTurnAudioArtifact {
+ audio_artifact: Some(RuntimeTurnAudioArtifact {
artifact_type: "local_file".to_string(),
path_ref: path_ref.to_string(),
format: "wav".to_string(),
- sample_rate: TARGET_SAMPLE_RATE_HZ,
- channels: u32::from(TARGET_NUM_CHANNELS),
+ sample_rate: USER_AUDIO_SAMPLE_RATE_HZ,
+ channels: u32::from(USER_AUDIO_NUM_CHANNELS),
duration_ms: turn.duration_ms,
byte_size,
- },
+ }),
+ asr_result_ref: asr_result_ref.map(str::to_string),
};
let response = http
.post(bridge_config.bridge_url.as_deref().unwrap_or_default())
@@ -1343,20 +1683,59 @@
.trim()
.to_ascii_lowercase();
let frames = if format == "pcm_s16le" {
- pcm_s16le_payload_to_frames(
- &payload,
- audio_chunk.sample_rate.unwrap_or(TARGET_SAMPLE_RATE_HZ),
- audio_chunk
- .channels
- .unwrap_or(u32::from(TARGET_NUM_CHANNELS)),
- )?
+ let sample_rate = audio_chunk.sample_rate.unwrap_or(sink.sample_rate_hz);
+ let channels = audio_chunk.channels.unwrap_or(u32::from(sink.num_channels));
+ if state.pcm_stream_decoder.is_none() {
+ state.pcm_stream_decoder = Some(audio::PcmS16leStreamDecoder::new(
+ sample_rate,
+ channels,
+ sink.sample_rate_hz,
+ sink.num_channels,
+ )?);
+ }
+ state.pcm_stream_network_chunk_count =
+ state.pcm_stream_network_chunk_count.saturating_add(1);
+ let stream_result = state
+ .pcm_stream_decoder
+ .as_mut()
+ .expect("pcm stream decoder initialized")
+ .push_bytes(
+ &payload,
+ sample_rate,
+ channels,
+ audio_chunk.last.unwrap_or(false),
+ )?;
+ if stream_result.dropped_tail_bytes > 0 {
+ warn!(
+ call_id = %call_id,
+ trace_id = %trace_id,
+ turn_id = %turn.turn_id,
+ chunk_seq = audio_chunk.chunk_seq.unwrap_or_default(),
+ dropped_tail_bytes = stream_result.dropped_tail_bytes,
+ "runtime helper stream_audio_pcm_unaligned_tail_dropped"
+ );
+ }
+ if stream_result.frames.is_empty() {
+ warn!(
+ call_id = %call_id,
+ trace_id = %trace_id,
+ turn_id = %turn.turn_id,
+ chunk_seq = audio_chunk.chunk_seq.unwrap_or_default(),
+ buffered_source_bytes = stream_result.buffered_source_bytes,
+ buffered_source_samples = stream_result.buffered_source_samples,
+ network_chunk_count = state.pcm_stream_network_chunk_count,
+ "runtime helper stream_audio_pcm_waiting_for_20ms_frame"
+ );
+ return Ok(0);
+ }
+ stream_result.frames
} else if matches!(format.as_str(), "mp3" | "mpeg" | "wav") {
state.encoded_audio_buffer.extend_from_slice(&payload);
match audio::decode_audio_bytes_to_frames(
&state.encoded_audio_buffer,
"stream_chunk",
- TARGET_SAMPLE_RATE_HZ,
- TARGET_NUM_CHANNELS,
+ sink.sample_rate_hz,
+ sink.num_channels,
bridge_config.audio_debug_dump_dir.as_deref(),
call_id,
&format!("stream-reply-{}", turn.turn_id),
@@ -1441,6 +1820,34 @@
sleep_until(pacing_started_at + Duration::from_millis(((index + 1) as u64) * 20)).await;
}
if audio_chunk.last.unwrap_or(false) {
+ let mut debug_source_path = None;
+ let mut debug_pcm_wav_path = None;
+ let mut debug_pcm_wav_size_bytes = None;
+ if format == "pcm_s16le" {
+ if let (Some(debug_dump_dir), Some(decoder)) = (
+ bridge_config.audio_debug_dump_dir.as_deref(),
+ state.pcm_stream_decoder.as_ref(),
+ ) {
+ match decoder.write_debug_dump(
+ debug_dump_dir,
+ call_id,
+ &format!("stream-reply-{}", turn.turn_id),
+ ) {
+ Ok(debug_dump) => {
+ debug_source_path = debug_dump.debug_source_path;
+ debug_pcm_wav_path = debug_dump.debug_pcm_wav_path;
+ debug_pcm_wav_size_bytes = debug_dump.debug_pcm_wav_size_bytes;
+ }
+ Err(error) => warn!(
+ call_id = %call_id,
+ trace_id = %trace_id,
+ turn_id = %turn.turn_id,
+ error = %safe_error(&error.to_string()),
+ "runtime helper stream_audio_pcm_debug_dump_failed"
+ ),
+ }
+ }
+ }
emit_activity(
call_id,
trace_id,
@@ -1453,25 +1860,20 @@
"replyPlaybackMode": state.reply_playback_mode.as_str(),
"format": format.as_str(),
"chunkSeq": audio_chunk.chunk_seq,
+ "networkChunkCount": state.pcm_stream_network_chunk_count,
+ "debugSourcePath": debug_source_path,
+ "debugPcmWavPath": debug_pcm_wav_path,
+ "debugPcmWavSizeBytes": debug_pcm_wav_size_bytes,
+ "sourceSampleRate": audio_chunk.sample_rate,
+ "sourceChannels": audio_chunk.channels,
+ "targetAudioProfile": sink.profile.as_str(),
+ "targetSampleRate": sink.sample_rate_hz,
+ "targetChannels": sink.num_channels,
"replyTotalAfterVadEndMs": turn_pipeline_started_at.elapsed().as_millis() as u64,
}),
);
}
Ok(frames.len())
-}
-
-fn pcm_s16le_payload_to_frames(
- payload: &[u8],
- sample_rate: u32,
- channels: u32,
-) -> Result<Vec<PcmFrame>> {
- audio::pcm_s16le_bytes_to_frames(
- payload,
- sample_rate,
- channels,
- TARGET_SAMPLE_RATE_HZ,
- TARGET_NUM_CHANNELS,
- )
}
fn trim_ascii_whitespace(value: &[u8]) -> &[u8] {
@@ -1494,6 +1896,7 @@
turn: &FinishedSpeechTurn,
path_ref: &str,
byte_size: u64,
+ asr_result_ref: Option<&str>,
turn_pipeline_started_at: Instant,
) -> Result<RuntimeTurnBridgeOutcome> {
let bridge_started_at = Instant::now();
@@ -1501,15 +1904,16 @@
call_id: call_id.to_string(),
trace_id: trace_id.to_string(),
turn_id: turn.turn_id.clone(),
- audio_artifact: RuntimeTurnAudioArtifact {
+ audio_artifact: Some(RuntimeTurnAudioArtifact {
artifact_type: "local_file".to_string(),
path_ref: path_ref.to_string(),
format: "wav".to_string(),
- sample_rate: TARGET_SAMPLE_RATE_HZ,
- channels: u32::from(TARGET_NUM_CHANNELS),
+ sample_rate: USER_AUDIO_SAMPLE_RATE_HZ,
+ channels: u32::from(USER_AUDIO_NUM_CHANNELS),
duration_ms: turn.duration_ms,
byte_size,
- },
+ }),
+ asr_result_ref: asr_result_ref.map(str::to_string),
};
let response = http
.post(bridge_config.bridge_url.as_deref().unwrap_or_default())
@@ -1669,8 +2073,8 @@
http,
Some(&audio_path_string),
None,
- TARGET_SAMPLE_RATE_HZ,
- TARGET_NUM_CHANNELS,
+ sink.sample_rate_hz,
+ sink.num_channels,
bridge_config.audio_debug_dump_dir.as_deref(),
call_id,
&reply_debug_label,
@@ -1800,7 +2204,10 @@
#[serde(rename = "turnId")]
turn_id: String,
#[serde(rename = "audioArtifact")]
- audio_artifact: RuntimeTurnAudioArtifact,
+ #[serde(skip_serializing_if = "Option::is_none")]
+ audio_artifact: Option<RuntimeTurnAudioArtifact>,
+ #[serde(rename = "asrResultRef", skip_serializing_if = "Option::is_none")]
+ asr_result_ref: Option<String>,
}
#[derive(Serialize)]
@@ -1828,6 +2235,20 @@
retryable: Option<bool>,
}
+#[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
+struct RuntimeAsrStreamResp {
+ status: Option<String>,
+ asr_result_ref: Option<String>,
+ chunk_count: Option<u64>,
+ audio_bytes: Option<u64>,
+ asr_duration_ms: Option<u64>,
+ provider_alias: Option<String>,
+ text_len: Option<u64>,
+ fallback_reason: Option<String>,
+ fallback_stage: Option<String>,
+}
+
#[derive(Default)]
struct RuntimeTurnBridgeOutcome {
reply_audio_artifact: Option<RuntimeTurnReplyAudioArtifact>,
@@ -1849,6 +2270,8 @@
audio_chunk_count: u64,
device_output_count: u64,
encoded_audio_buffer: Vec<u8>,
+ pcm_stream_decoder: Option<audio::PcmS16leStreamDecoder>,
+ pcm_stream_network_chunk_count: u64,
}
impl Default for RuntimeTurnStreamState {
@@ -1862,6 +2285,8 @@
audio_chunk_count: 0,
device_output_count: 0,
encoded_audio_buffer: Vec::new(),
+ pcm_stream_decoder: None,
+ pcm_stream_network_chunk_count: 0,
}
}
}
@@ -2010,8 +2435,8 @@
tokio::spawn(async move {
let mut stream = NativeAudioStream::new(
track.rtc_track(),
- TARGET_SAMPLE_RATE_HZ as i32,
- i32::from(TARGET_NUM_CHANNELS),
+ USER_AUDIO_SAMPLE_RATE_HZ as i32,
+ i32::from(USER_AUDIO_NUM_CHANNELS),
);
let started_at = Instant::now();
let mut frame_count: u64 = 0;
@@ -2128,8 +2553,8 @@
rms_threshold: f64_env("CV_VAD_RMS_THRESHOLD", 0.012),
peak_threshold: f64_env("CV_VAD_PEAK_THRESHOLD", 0.08),
start_frames: u32_env("CV_VAD_START_FRAMES", 5).max(1),
- end_silence_ms: u64_env("CV_VAD_END_SILENCE_MS", 700).max(100),
- min_speech_ms: u64_env("CV_VAD_MIN_SPEECH_MS", 300).max(1),
+ end_silence_ms: u64_env("CV_VAD_END_SILENCE_MS", 400).max(100),
+ min_speech_ms: u64_env("CV_VAD_MIN_SPEECH_MS", 250).max(1),
max_turn_ms: u64_env("CV_VAD_MAX_TURN_MS", 10_000).max(1_000),
initial_ignore_ms: u64_env("CV_VAD_INITIAL_IGNORE_MS", 500),
}
@@ -2242,7 +2667,9 @@
return None;
}
- let voiced = rms >= self.config.rms_threshold || peak >= self.config.peak_threshold;
+ // Align with cb-sdk's energy-based segmentation: peak is diagnostic only,
+ // otherwise isolated spikes can keep a turn open until max_turn_ms.
+ let voiced = rms >= self.config.rms_threshold;
if !self.in_speech {
self.remember_pre_speech_frame(frame);
}
@@ -2498,6 +2925,9 @@
rtc_source: NativeAudioSource,
track: LocalAudioTrack,
device_output_destination_identity: Option<String>,
+ profile: String,
+ sample_rate_hz: u32,
+ num_channels: u16,
}
impl BotAudioOutputSink {
@@ -2508,6 +2938,7 @@
call_id: &str,
trace_id: &str,
track_name: &str,
+ profile: String,
sample_rate: u32,
num_channels: u32,
device_output_destination_identity: Option<String>,
@@ -2536,11 +2967,14 @@
"failed to publish bot audio track in room {room_alias} for participant {participant_alias}: {error}"
)
})?;
+ let num_channels_u16 = u16::try_from(num_channels)
+ .map_err(|_| anyhow!("unsupported bot audio channel count {num_channels}"))?;
info!(
room_alias = %room_alias,
participant_alias = %participant_alias,
track_name = %track_name,
+ bot_audio_profile = %profile,
sample_rate,
num_channels,
"runtime helper published bot audio track"
@@ -2555,6 +2989,7 @@
None,
json!({
"trackName": track_name,
+ "audioProfile": profile,
"sampleRate": sample_rate,
"numChannels": num_channels,
}),
@@ -2565,6 +3000,9 @@
rtc_source,
track,
device_output_destination_identity,
+ profile,
+ sample_rate_hz: sample_rate,
+ num_channels: num_channels_u16,
})
}
--
Gitblit v1.9.3