tirtc-devtools-cli 0.3.1 → 0.3.3

package/dist/cli/src/index.js

@@ -125,6 +125,10 @@ device.command('start')
     .option('--endpoint <url>', 'TiRTC endpoint；不传时读取 TIRTC_ENDPOINT')
     .option('--source <path>', 'prepared asset root、manifest_path 或 encoded track；MP4 先运行 assets prepare')
     .option('--video-codec <codec>', 'h264|h265|mjpeg', 'h264')
+    .option('--audio-codec <codec>', 'g711a|aac', 'g711a')
+    .option('--audio-sample-rate <hz>', '8000|16000', '8000')
+    .option('--audio-channels <count>', '1|2', '1')
+    .option('--exit-after-first-session', '首个 client 会话完成后主动正常退出并写出 summary')
     .option('--duration-ms <ms>', '可选自动结束时长；默认持续运行直到用户结束进程')
     .option('--connect-timeout-ms <ms>', 'service ready 最大等待')
     .option('--first-packet-timeout-ms <ms>', '首包最大等待')

package/dist/cli/src/role_driver.d.ts

@@ -8,6 +8,10 @@ type DeviceCommandOptions = {
     endpoint?: string;
     source?: string;
     videoCodec?: string;
+    audioCodec?: string;
+    audioSampleRate?: string;
+    audioChannels?: string;
+    exitAfterFirstSession?: boolean;
     durationMs?: string;
     connectTimeoutMs?: string;
     firstPacketTimeoutMs?: string;

package/dist/cli/src/role_driver.js

@@ -39,6 +39,9 @@ function rolePreflightError(reasonCode, detail) {
 function roleUsageError(message) {
     return new RoleCommandError('invalid_request', 'config', usageExitCode, message);
 }
+function roleUsageReasonError(reasonCode, message) {
+    return new RoleCommandError(reasonCode, 'config', usageExitCode, reasonCode + ': ' + message);
+}
 function pathExists(filePath) {
     return fs_1.default.existsSync(filePath);
 }
@@ -270,6 +273,27 @@ function codecOrDefault(raw) {
     }
     return codec;
 }
+function audioCodecOrDefault(raw) {
+    const codec = raw?.trim() || 'g711a';
+    if (codec !== 'g711a' && codec !== 'aac') {
+        throw roleUsageReasonError('audio_codec_unsupported', 'audio-codec must be g711a or aac');
+    }
+    return codec;
+}
+function audioSampleRateOrDefault(raw) {
+    const sampleRate = raw === undefined ? 8000 : Number(raw);
+    if (sampleRate !== 8000 && sampleRate !== 16000) {
+        throw roleUsageReasonError('audio_format_unsupported', 'audio-sample-rate must be 8000 or 16000');
+    }
+    return sampleRate;
+}
+function audioChannelsOrDefault(raw) {
+    const channels = raw === undefined ? 1 : Number(raw);
+    if (channels !== 1 && channels !== 2) {
+        throw roleUsageReasonError('audio_format_unsupported', 'audio-channels must be 1 or 2');
+    }
+    return channels;
+}
 function consumerOrDefault(raw) {
     const consumer = raw?.trim() || 'frame_dump';
     if (consumer !== 'frame_dump' && consumer !== 'packet_dump') {
@@ -302,6 +326,9 @@ function printEnvelope(options, code, message, data) {
 }
 function buildDeviceRequest(roots, artifactRoot, options) {
     const codec = codecOrDefault(options.videoCodec);
+    const audioCodec = audioCodecOrDefault(options.audioCodec);
+    const audioSampleRateHz = audioSampleRateOrDefault(options.audioSampleRate);
+    const audioChannels = audioChannelsOrDefault(options.audioChannels);
     const executionId = 'cli-device-' + codec + '-' + executionSuffix();
     const caseId = 'devtools-cli-device.' + codec;
     const deviceIdentity = resolveDeviceIdentity(options);
@@ -328,9 +355,15 @@ function buildDeviceRequest(roots, artifactRoot, options) {
         media: {
             source: { kind: 'encoded_asset', path: options.source ?? resolveAssetRoot(roots) },
             video: { codec },
+            audio: {
+                codec: audioCodec,
+                sample_rate_hz: audioSampleRateHz,
+                channels: audioChannels,
+            },
         },
         output: { consumer: 'frame_dump', video: { frame_limit: defaultFrameLimit } },
         run: {
+            exit_after_first_session: options.exitAfterFirstSession === true,
             duration_ms: parseOptionalPositiveInt(options.durationMs, '--duration-ms'),
             connect_timeout_ms: parsePositiveInt(options.connectTimeoutMs, defaultConnectTimeoutMs, '--connect-timeout-ms'),
             first_packet_timeout_ms: parsePositiveInt(options.firstPacketTimeoutMs, defaultFirstPacketTimeoutMs, '--first-packet-timeout-ms'),
@@ -372,11 +405,15 @@ function buildClientRequest(roots, artifactRoot, options) {
     const executionId = 'cli-client-' + executionSuffix();
     const identityOverride = Boolean(bootstrap && (options.targetDeviceId || options.token));
     const videoCodec = codecOrDefault(bootstrap?.video_codec);
+    const audioCodec = audioCodecOrDefault(bootstrap?.audio_codec);
+    const audioSampleRateHz = audioSampleRateOrDefault(bootstrap?.audio_sample_rate_hz ?? bootstrap?.sample_rate_hz);
+    const audioChannels = audioChannelsOrDefault(bootstrap?.audio_channels ?? bootstrap?.channels);
     return {
         schema_version: 1,
         execution_id: executionId,
         case_id: 'devtools-cli-client',
         role: 'client',
+        require_audio: bootstrap?.require_audio ?? true,
         endpoint,
         identity: {
             device_id: targetDeviceId,
@@ -392,6 +429,11 @@ function buildClientRequest(roots, artifactRoot, options) {
         media: {
             source: { kind: 'encoded_asset', path: resolveAssetRoot(roots) },
             video: { codec: videoCodec },
+            audio: {
+                codec: audioCodec,
+                sample_rate_hz: audioSampleRateHz,
+                channels: audioChannels,
+            },
         },
         output: {
             consumer: consumerOrDefault(options.consumer),

package/dist/cli/src/role_live_log.js

@@ -166,7 +166,7 @@ function startRoleLiveLog(options) {
                 const parts = ['audio input started'];
                 appendField(parts, 'session', sessionIndex(payload));
                 appendField(parts, 'stream_id', payloadNumber(payload, 'stream_id'));
-                appendField(parts, 'codec', payloadString(payload, 'codec'));
+                appendField(parts, 'codec', payloadString(payload, 'audio_codec') ?? payloadString(payload, 'codec'));
                 appendField(parts, 'sample_rate_hz', payloadNumber(payload, 'sample_rate_hz'));
                 log(parts.join(' '));
                 break;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tirtc-devtools-cli",
-  "version": "0.3.1",
+  "version": "0.3.3",
   "private": false,
   "main": "dist/cli/src/index.js",
   "types": "dist/cli/src/index.d.ts",

package/vendor/runtime/linux-x64/include/tirtc/audio.h

@@ -42,7 +42,11 @@ typedef enum TirtcAudioOutputState {
 } TirtcAudioOutputState;
 
 typedef struct TirtcAudioInputOptions {
-  TirtcAudioSampleRate sample_rate;
+  TirtcMediaCodec codec;
+
+  uint32_t sample_rate_hz;
+
+  uint32_t channels;
 
   TirtcAudioAecMode aec_mode;
 
@@ -61,9 +65,27 @@ typedef struct TirtcAudioInputOptions {
 typedef struct TirtcAudioEncodedInputOptions {
   TirtcMediaCodec codec;
 
-  TirtcAudioSampleRate sample_rate;
+  uint32_t sample_rate_hz;
+
+  uint32_t channels;
 } TirtcAudioEncodedInputOptions;
 
+typedef struct TirtcAudioEncodedInputFrame {
+  TirtcMediaCodec codec;
+
+  uint32_t sample_rate_hz;
+
+  uint32_t channels;
+
+  uint32_t samples_per_channel;
+
+  int64_t pts_us;
+
+  const void* data;
+
+  size_t data_bytes;
+} TirtcAudioEncodedInputFrame;
+
 typedef struct TirtcAudioOutputOptions {
   uint32_t volume_percent;
 
@@ -91,6 +113,22 @@ typedef struct TirtcPendingDurationMetrics {
   uint64_t decoded_duration_ms;
 } TirtcPendingDurationMetrics;
 
+typedef struct TirtcOutputLatencyStageMetrics {
+  uint64_t latest_ms;
+  uint64_t average_ms;
+  uint64_t peak_ms;
+  uint64_t sample_count;
+  uint64_t unavailable_count;
+} TirtcOutputLatencyStageMetrics;
+
+typedef struct TirtcOutputLocalLatencyMetrics {
+  uint32_t window_duration_ms;
+  TirtcOutputLatencyStageMetrics total;
+  TirtcOutputLatencyStageMetrics buffer;
+  TirtcOutputLatencyStageMetrics decode_or_ready;
+  TirtcOutputLatencyStageMetrics output;
+} TirtcOutputLocalLatencyMetrics;
+
 typedef struct TirtcAudioOutputMetricsSnapshot {
   TirtcOutputStutterMetrics stutter;
   TirtcPendingDurationMetrics pending;
@@ -98,6 +136,7 @@ typedef struct TirtcAudioOutputMetricsSnapshot {
   double input_packet_rate;
   double render_callback_rate;
   uint32_t rate_window_duration_ms;
+  TirtcOutputLocalLatencyMetrics local_latency;
 } TirtcAudioOutputMetricsSnapshot;
 
 /**
@@ -109,6 +148,10 @@ typedef struct TirtcAudioOutputMetricsSnapshot {
  */
 typedef struct TirtcAudioOutputDebugSnapshot {
   TirtcMediaCodec codec;
+
+  uint32_t sample_rate_hz;
+
+  uint32_t channels;
 } TirtcAudioOutputDebugSnapshot;
 
 /**
@@ -195,7 +238,7 @@ TirtcError tirtc_audio_encoded_input_attach(TirtcAudioEncodedInput* input, Tirtc
 TirtcError tirtc_audio_encoded_input_detach(TirtcAudioEncodedInput* input, TirtcConn* connection);
 
 TirtcError tirtc_audio_encoded_input_submit_frame(TirtcAudioEncodedInput* input,
-                                                  const TirtcAudioEncodedFrame* frame);
+                                                  const TirtcAudioEncodedInputFrame* frame);
 
 void tirtc_audio_encoded_input_destroy(TirtcAudioEncodedInput* input);
 

package/vendor/runtime/linux-x64/include/tirtc/audio_codec.h

@@ -16,6 +16,12 @@ TirtcError tirtc_audio_g711a_encode(const TirtcAudioPcmFrame* pcm_frame, void* o
 TirtcError tirtc_audio_g711a_decode(const TirtcAudioEncodedFrame* in_frame, void* out_buffer,
                                     size_t out_buffer_capacity, TirtcAudioPcmFrame* out_frame);
 
+TirtcError tirtc_audio_aac_encode(const TirtcAudioPcmFrame* pcm_frame, void* out_buffer,
+                                  size_t out_buffer_capacity, TirtcAudioEncodedFrame* out_frame);
+
+TirtcError tirtc_audio_aac_decode(const TirtcAudioEncodedFrame* in_frame, void* out_buffer,
+                                  size_t out_buffer_capacity, TirtcAudioPcmFrame* out_frame);
+
 #ifdef __cplusplus
 }
 #endif

package/vendor/runtime/linux-x64/include/tirtc/audio_format.h

@@ -0,0 +1,20 @@
+#ifndef TIRTC_MEDIA_CONTRACT_AUDIO_FORMAT_H_
+#define TIRTC_MEDIA_CONTRACT_AUDIO_FORMAT_H_
+
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct TirtcAudioFormat {
+  uint32_t sample_rate_hz;
+  uint32_t channels;
+  uint32_t bits_per_sample;
+} TirtcAudioFormat;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // TIRTC_MEDIA_CONTRACT_AUDIO_FORMAT_H_

package/vendor/runtime/linux-x64/include/tirtc/audio_frame.h

@@ -4,7 +4,7 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#include "tirtc/audio_sample_rate.h"
+#include "tirtc/audio_format.h"
 #include "tirtc/media_codec.h"
 
 #ifdef __cplusplus
@@ -12,17 +12,16 @@ extern "C" {
 #endif
 
 typedef struct TirtcAudioPcmFrame {
-  TirtcAudioSampleRate sample_rate;
+  TirtcAudioFormat format;
   uint32_t samples_per_channel;
   int64_t pts_us;
-  /* Mutable PCM payload for capture, processing, and render-fill paths. */
   void* data;
   size_t data_bytes;
 } TirtcAudioPcmFrame;
 
 typedef struct TirtcAudioEncodedFrame {
   TirtcMediaCodec codec;
-  TirtcAudioSampleRate sample_rate;
+  TirtcAudioFormat format;
   uint32_t samples_per_channel;
   int64_t pts_us;
   const void* data;

package/vendor/runtime/linux-x64/include/tirtc/audio_io.h

@@ -14,8 +14,7 @@ typedef struct TirtcAudioAin TirtcAudioAin;
 typedef struct TirtcAudioAout TirtcAudioAout;
 
 typedef struct TirtcAudioIoConfig {
-  TirtcAudioSampleRate sample_rate;
-  /* AIN uses fixed 10 ms mono blocks; AOUT may use larger callback-sized multiples. */
+  TirtcAudioFormat format;
   uint32_t samples_per_buffer;
 } TirtcAudioIoConfig;
 
@@ -29,7 +28,7 @@ typedef void (*TirtcAudioAinOnFrameFn)(TirtcAudioAin* ain, const TirtcAudioPcmFr
                                        void* user_data);
 
 /*
- * AOUT pre-fills sample_rate, samples_per_channel, data, and data_bytes before
+ * AOUT pre-fills format, samples_per_channel, data, and data_bytes before
  * calling render. The callback contract expects a non-negative rendered PCM
  * byte count. Backends defensively clamp negative returns to zero and zero-fill
  * any remaining buffer space.

package/vendor/runtime/linux-x64/include/tirtc/audio_processing.h

@@ -2,6 +2,7 @@
 #define TIRTC_AUDIO_PROCESSING_H_
 
 #include "tirtc/audio_frame.h"
+#include "tirtc/audio_sample_rate.h"
 #include "tirtc/error.h"
 
 #ifdef __cplusplus

package/vendor/runtime/linux-x64/include/tirtc/av.h

@@ -51,6 +51,7 @@ typedef struct TirtcVideoOutputMetricsSnapshot {
   double decoded_fps;
   double render_fps;
   uint32_t rate_window_duration_ms;
+  TirtcOutputLocalLatencyMetrics local_latency;
 } TirtcVideoOutputMetricsSnapshot;
 
 /**

package/vendor/runtime/linux-x64/include/tirtc/media_codec.h

@@ -8,6 +8,7 @@ extern "C" {
 typedef enum TirtcMediaCodec {
   TIRTC_MEDIA_CODEC_NONE = 0,
   TIRTC_MEDIA_CODEC_AUDIO_G711A = 1,
+  TIRTC_MEDIA_CODEC_AUDIO_AAC = 2,
   /* Reserved shared identifiers; concrete video implementations live outside media-contract. */
   TIRTC_MEDIA_CODEC_VIDEO_H264 = 65,
   TIRTC_MEDIA_CODEC_VIDEO_H265 = 66,

package/vendor/runtime/linux-x64/include/tirtc/media_downlink.h

@@ -66,6 +66,22 @@ typedef struct TirtcMediaDownlinkPendingMetricsSnapshot {
   uint64_t decoded_pending_duration_ms;
 } TirtcMediaDownlinkPendingMetricsSnapshot;
 
+typedef struct TirtcMediaDownlinkLatencyStageMetricsSnapshot {
+  uint64_t latest_ms;
+  uint64_t average_ms;
+  uint64_t peak_ms;
+  uint64_t sample_count;
+  uint64_t unavailable_count;
+} TirtcMediaDownlinkLatencyStageMetricsSnapshot;
+
+typedef struct TirtcMediaDownlinkLocalLatencyMetricsSnapshot {
+  uint32_t window_duration_ms;
+  TirtcMediaDownlinkLatencyStageMetricsSnapshot total;
+  TirtcMediaDownlinkLatencyStageMetricsSnapshot buffer;
+  TirtcMediaDownlinkLatencyStageMetricsSnapshot decode_or_ready;
+  TirtcMediaDownlinkLatencyStageMetricsSnapshot output;
+} TirtcMediaDownlinkLocalLatencyMetricsSnapshot;
+
 typedef struct TirtcMediaDownlinkAudioMetricsSnapshot {
   TirtcMediaDownlinkStutterMetricsSnapshot stutter;
   TirtcMediaDownlinkPendingMetricsSnapshot pending;
@@ -73,6 +89,7 @@ typedef struct TirtcMediaDownlinkAudioMetricsSnapshot {
   double input_packet_rate;
   double render_callback_rate;
   uint32_t rate_window_duration_ms;
+  TirtcMediaDownlinkLocalLatencyMetricsSnapshot local_latency;
 } TirtcMediaDownlinkAudioMetricsSnapshot;
 
 typedef struct TirtcMediaDownlinkVideoStartupMetricsSnapshot {
@@ -93,6 +110,7 @@ typedef struct TirtcMediaDownlinkVideoMetricsSnapshot {
   double decoded_fps;
   double render_fps;
   uint32_t rate_window_duration_ms;
+  TirtcMediaDownlinkLocalLatencyMetricsSnapshot local_latency;
 } TirtcMediaDownlinkVideoMetricsSnapshot;
 
 /**
@@ -103,6 +121,7 @@ typedef struct TirtcMediaDownlinkVideoMetricsSnapshot {
  */
 typedef struct TirtcMediaDownlinkAudioDebugSnapshot {
   TirtcMediaCodec codec;
+  TirtcAudioFormat format;
 } TirtcMediaDownlinkAudioDebugSnapshot;
 
 /**

package/vendor/runtime/linux-x64/include/tirtc/media_uplink.h

@@ -29,7 +29,8 @@ typedef enum TirtcMediaUplinkState {
 } TirtcMediaUplinkState;
 
 typedef struct TirtcMediaUplinkAudioOptions {
-  TirtcAudioSampleRate sample_rate;
+  TirtcMediaCodec codec;
+  TirtcAudioFormat format;
   TirtcAudioAecMode aec_mode;
   TirtcAudioAgcLevel agc_level;
   TirtcAudioAnsLevel ans_level;

package/vendor/runtime/linux-x64/manifest.txt

@@ -1,26 +1,27 @@
 platform=linux-x64
 profile=full
-staged_at_utc=2026-05-09T08:58:54Z
+staged_at_utc=2026-05-13T16:55:41Z
 source_sdk=/Users/allenfeng/Development/Repositories/tirtc-nexus/tirtc-matrix/.build/sdk/linux-x64
 
-92ac54ea714e5a7da536d3d63b955a9e30faf9560ebe670034a79f3ea69c21cb  include/tirtc/audio.h
-6d972ccfe150a3b4f2d7f18fa92b3ade9210c1c8bb754d061ac6b7997b59e2cb  include/tirtc/audio_codec.h
-7bacbdb2d8bb10d6444036a8fef42f2a8e3ea34dfc38e165ee678d61f189db41  include/tirtc/audio_frame.h
-fb8ea35263167c33487bd3aa503c26ed2dfbf97877f7bda92c10065cd5c97c13  include/tirtc/audio_io.h
+04c3139f26d53aa2fb0a586a21ffb481a001c01e0412cd3414a952c9a9d336ba  include/tirtc/audio.h
+ce7ccffa054250a2dab0e2128e4701a66c7168e50e8c470dcff3bf609ff0959c  include/tirtc/audio_codec.h
+0b463d8adfcf45acb3f95fe3e2d94b37a26aaf37d293eb2b18274ddcb161f767  include/tirtc/audio_format.h
+7cac5b8c0cef28526ee00800fa8b0d1e60897ab971a0801899cb0333c9189fde  include/tirtc/audio_frame.h
+6daadba731521c01a3cf11c5d8b2ef797238737c939e2da3045fc83ad89af04a  include/tirtc/audio_io.h
 c2e1f31dcc75be461c577d18b1cebe32774f212d51cb4dd2a5b5a9bfe62b693e  include/tirtc/audio_io_android.h
 878bb18220ac62304ec95807f31e83e050c8dac14811f873454f9b045028694a  include/tirtc/audio_io_apple.h
 0569eb1d043ee7fa8d45ca43143eca1a22e4c4bf1507ca7e045e010a3ecbc288  include/tirtc/audio_io_harmony.h
 51cbc911fe9f9834046f0e0a1a7cdd814a8e194a615894a8b4d11f9e5f095610  include/tirtc/audio_io_windows.h
-21f60729117260a44af22c1af986ef17d22673b102b7b7a035f492d0665cce16  include/tirtc/audio_processing.h
+5acfd19fdcd8c9d96e4aee7c6f1d33a9211c0a5b0c0ab56d96e1b4dee48f3aa9  include/tirtc/audio_processing.h
 0ca7c3c630b1242f51a0fd8154097c0a332b4c816a5707090e4381719852998c  include/tirtc/audio_sample_rate.h
-f40ce007241e4aa6ef2e2bcd589060636ef0e9859a9102c8b955d348f4f7eabc  include/tirtc/av.h
+07ca1b210a6e334c1474f25446477681472154568ecf74557cb3dc9b3a730662  include/tirtc/av.h
 b3e110ce04057d3be696b120d1abeaf6141711a1c63ca7f9bdf65ed5ec6e16d7  include/tirtc/error.h
 ae805545a9515edc9b94262e72ad2c7b7d649288166f4daeb450d8a55e82ae0b  include/tirtc/foundation/build_info.h
 7cf8b372a3d48d4de4a65a04c7f102281a7b42cebb9ec247853d3c53afb63b6a  include/tirtc/http.h
 70bbf93b84d9d1a85f376d9986de570c1f658319e1e5ab6d621f7a4d41033f5c  include/tirtc/logging.h
-7e2f7f81afe37e22f5414d9db32438edd948ff775230c3a0a17ff9ac1a0b560e  include/tirtc/media_codec.h
-43c0ac9173b6259a1b1379eac780f4892a0a72572801ac9eb8516c0bdb4913c3  include/tirtc/media_downlink.h
-e64191a39004165cdcd11b2ca824af34d4bfe02f7ec3c3101fa9291f56488c30  include/tirtc/media_uplink.h
+1419ac3fb5237d1d820bd30a6845b98aab3553463eb64e73623b64c663665f16  include/tirtc/media_codec.h
+df11627da9bfbb87cb32582701f638d2a0ff5b6b77b49a3a3bab7509c62f0613  include/tirtc/media_downlink.h
+9376af63fde915753146969291f0fefe2da6df28f7aa6310f7787048165150dc  include/tirtc/media_uplink.h
 57be6941026e7dd173e8779b2bb868532d65e41f7917aad476a31ac25c1cd6c3  include/tirtc/transport.h
 fba7954919606e507b6a074440440903665217959699da89ca9dd206772c0ef0  include/tirtc/trp.h
 92b3db6e2cb02265ad5db01881edce9d4e095b1dadf449dce5d8bb5362957e33  include/tirtc/video_codec.h
@@ -31,15 +32,18 @@ f22127424d5978ab0644209b6c25c24e19e320dd2fbda6634e6344597ff7e2be  include/tirtc/
 cae0bbeb884e5466a56da15182c78cc22baab6c743f349a58d3595f623333585  include/tirtc/video_io_harmony.h
 65b5d24fe3a6bc2a299a2220148a153ba4eb0955a49c1c3baabf1fdfed8aabdd  include/tirtc/video_io_windows.h
 8cd6b66bea14890a665cc317f8572429b2c3e4463773f8b77c1e4dc30a4a8747  include/tirtc/video_processing.h
-ceb7cdd45efae57e5bd53d4177cb20c4a0cb9917fcdd23127b050d69c8b919ac  lib/libTiRTC.a
+82653324cc9b6ea732ba8a4818175b8f99a5748ac0fa89f7e5c5e366a7f9e142  lib/libTiRTC.a
 b1f4135025cb8e8520a14268d831b9998920a4239f84bb6b409d0e4909fd5bee  lib/libcrypto.a
-296ba0b5efa9d2ac49e3bf278c5d467b31436e73fbc608b736e974bcf9759bde  lib/libmatrix_runtime_audio.a
-8552aba878ffb4c242e4af589784fafbdd938291663af750aaa94d902cba72ce  lib/libmatrix_runtime_facade.a
+8a3b5da754707a9c6fbb73ab6176cdb12bc8af22d3a660249cb1812d501fb615  lib/libmatrix_runtime_audio.a
+76a97c258a718c79846aadc91a389a5bf411de95bb43c0bd9b8a3ed3a2e8fa58  lib/libmatrix_runtime_facade.a
 ec7bbf8e8744c6645a441a28e24e0645906152cc987d6fcca16af3c3badba940  lib/libmatrix_runtime_foundation_http.a
-b4f225727bf9c829c3329054112ab56477f1cc5729be1eb4952a85b649ba0a8a  lib/libmatrix_runtime_foundation_logging.a
-031927b01cee74366cf2662559c2ddfe8476531c4af7c549f311503d4da6f424  lib/libmatrix_runtime_media.a
-4942f5c742a6013e35708f6dd1e5b8ca702454990f1785e82d734aaea6ab114d  lib/libmatrix_runtime_transport.a
-774e8ae6c5e96ff5976bf440bc259044157c51e0b1dab0e77e5e087b7d059f01  lib/libmatrix_runtime_video.a
+a5759ecf8887d249209ba358a548e6878373a6854991299725d5fd4ed77ed2ed  lib/libmatrix_runtime_foundation_logging.a
+df52cb3856c179698761357ef40c063dcdb86fa67c0be186f64dcc9c634c36b4  lib/libmatrix_runtime_media.a
+177e9ca89640bca3ae49a164e820b7e33c0d22d14dcd7458c358e74a47ffbd70  lib/libmatrix_runtime_transport.a
+e00ea11ed72e1a2414251ec9f5b3dea46ed440a5fe30f8d45ae64e60b1d0ac19  lib/libmatrix_runtime_video.a
+4b4b3170a775282fe2107f615539c985b4df43315572796faba23886aab62b1d  lib/libmbedtls.a
 511d521f7972df3993e5976d6e2dbcad7a6fbce9be15071274d0cbd9d51157f5  lib/libssl.a
+c3adc781b817481c5546e4a9bdd863dd192d204b3502a0de44a730793d4f6e64  lib/libusrsctp.a
+787f0d1f44f6f755305d1387172668a480c0ba9fdfdfb06af100351f957f6e38  lib/libwebrtc.a
 a67ec9034848ef24a1b17671e444daa62a8f9e6b8319a1e932593908720ff2a1  lib/libwebrtc_apm.a
 700e455255897e3cffab13ca593a6e4d9d11383ae609215cbbd6043a63d47161  lib/libxlog.a

package/vendor/runtime/macos-arm64/include/tirtc/audio.h

@@ -42,7 +42,11 @@ typedef enum TirtcAudioOutputState {
 } TirtcAudioOutputState;
 
 typedef struct TirtcAudioInputOptions {
-  TirtcAudioSampleRate sample_rate;
+  TirtcMediaCodec codec;
+
+  uint32_t sample_rate_hz;
+
+  uint32_t channels;
 
   TirtcAudioAecMode aec_mode;
 
@@ -61,9 +65,27 @@ typedef struct TirtcAudioInputOptions {
 typedef struct TirtcAudioEncodedInputOptions {
   TirtcMediaCodec codec;
 
-  TirtcAudioSampleRate sample_rate;
+  uint32_t sample_rate_hz;
+
+  uint32_t channels;
 } TirtcAudioEncodedInputOptions;
 
+typedef struct TirtcAudioEncodedInputFrame {
+  TirtcMediaCodec codec;
+
+  uint32_t sample_rate_hz;
+
+  uint32_t channels;
+
+  uint32_t samples_per_channel;
+
+  int64_t pts_us;
+
+  const void* data;
+
+  size_t data_bytes;
+} TirtcAudioEncodedInputFrame;
+
 typedef struct TirtcAudioOutputOptions {
   uint32_t volume_percent;
 
@@ -91,6 +113,22 @@ typedef struct TirtcPendingDurationMetrics {
   uint64_t decoded_duration_ms;
 } TirtcPendingDurationMetrics;
 
+typedef struct TirtcOutputLatencyStageMetrics {
+  uint64_t latest_ms;
+  uint64_t average_ms;
+  uint64_t peak_ms;
+  uint64_t sample_count;
+  uint64_t unavailable_count;
+} TirtcOutputLatencyStageMetrics;
+
+typedef struct TirtcOutputLocalLatencyMetrics {
+  uint32_t window_duration_ms;
+  TirtcOutputLatencyStageMetrics total;
+  TirtcOutputLatencyStageMetrics buffer;
+  TirtcOutputLatencyStageMetrics decode_or_ready;
+  TirtcOutputLatencyStageMetrics output;
+} TirtcOutputLocalLatencyMetrics;
+
 typedef struct TirtcAudioOutputMetricsSnapshot {
   TirtcOutputStutterMetrics stutter;
   TirtcPendingDurationMetrics pending;
@@ -98,6 +136,7 @@ typedef struct TirtcAudioOutputMetricsSnapshot {
   double input_packet_rate;
   double render_callback_rate;
   uint32_t rate_window_duration_ms;
+  TirtcOutputLocalLatencyMetrics local_latency;
 } TirtcAudioOutputMetricsSnapshot;
 
 /**
@@ -109,6 +148,10 @@ typedef struct TirtcAudioOutputMetricsSnapshot {
  */
 typedef struct TirtcAudioOutputDebugSnapshot {
   TirtcMediaCodec codec;
+
+  uint32_t sample_rate_hz;
+
+  uint32_t channels;
 } TirtcAudioOutputDebugSnapshot;
 
 /**
@@ -195,7 +238,7 @@ TirtcError tirtc_audio_encoded_input_attach(TirtcAudioEncodedInput* input, Tirtc
 TirtcError tirtc_audio_encoded_input_detach(TirtcAudioEncodedInput* input, TirtcConn* connection);
 
 TirtcError tirtc_audio_encoded_input_submit_frame(TirtcAudioEncodedInput* input,
-                                                  const TirtcAudioEncodedFrame* frame);
+                                                  const TirtcAudioEncodedInputFrame* frame);
 
 void tirtc_audio_encoded_input_destroy(TirtcAudioEncodedInput* input);
 

package/vendor/runtime/macos-arm64/include/tirtc/audio_codec.h

@@ -16,6 +16,12 @@ TirtcError tirtc_audio_g711a_encode(const TirtcAudioPcmFrame* pcm_frame, void* o
 TirtcError tirtc_audio_g711a_decode(const TirtcAudioEncodedFrame* in_frame, void* out_buffer,
                                     size_t out_buffer_capacity, TirtcAudioPcmFrame* out_frame);
 
+TirtcError tirtc_audio_aac_encode(const TirtcAudioPcmFrame* pcm_frame, void* out_buffer,
+                                  size_t out_buffer_capacity, TirtcAudioEncodedFrame* out_frame);
+
+TirtcError tirtc_audio_aac_decode(const TirtcAudioEncodedFrame* in_frame, void* out_buffer,
+                                  size_t out_buffer_capacity, TirtcAudioPcmFrame* out_frame);
+
 #ifdef __cplusplus
 }
 #endif

package/vendor/runtime/macos-arm64/include/tirtc/audio_format.h

@@ -0,0 +1,20 @@
+#ifndef TIRTC_MEDIA_CONTRACT_AUDIO_FORMAT_H_
+#define TIRTC_MEDIA_CONTRACT_AUDIO_FORMAT_H_
+
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct TirtcAudioFormat {
+  uint32_t sample_rate_hz;
+  uint32_t channels;
+  uint32_t bits_per_sample;
+} TirtcAudioFormat;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // TIRTC_MEDIA_CONTRACT_AUDIO_FORMAT_H_

package/vendor/runtime/macos-arm64/include/tirtc/audio_frame.h

@@ -4,7 +4,7 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#include "tirtc/audio_sample_rate.h"
+#include "tirtc/audio_format.h"
 #include "tirtc/media_codec.h"
 
 #ifdef __cplusplus
@@ -12,17 +12,16 @@ extern "C" {
 #endif
 
 typedef struct TirtcAudioPcmFrame {
-  TirtcAudioSampleRate sample_rate;
+  TirtcAudioFormat format;
   uint32_t samples_per_channel;
   int64_t pts_us;
-  /* Mutable PCM payload for capture, processing, and render-fill paths. */
   void* data;
   size_t data_bytes;
 } TirtcAudioPcmFrame;
 
 typedef struct TirtcAudioEncodedFrame {
   TirtcMediaCodec codec;
-  TirtcAudioSampleRate sample_rate;
+  TirtcAudioFormat format;
   uint32_t samples_per_channel;
   int64_t pts_us;
   const void* data;

package/vendor/runtime/macos-arm64/include/tirtc/audio_io.h

@@ -14,8 +14,7 @@ typedef struct TirtcAudioAin TirtcAudioAin;
 typedef struct TirtcAudioAout TirtcAudioAout;
 
 typedef struct TirtcAudioIoConfig {
-  TirtcAudioSampleRate sample_rate;
-  /* AIN uses fixed 10 ms mono blocks; AOUT may use larger callback-sized multiples. */
+  TirtcAudioFormat format;
   uint32_t samples_per_buffer;
 } TirtcAudioIoConfig;
 
@@ -29,7 +28,7 @@ typedef void (*TirtcAudioAinOnFrameFn)(TirtcAudioAin* ain, const TirtcAudioPcmFr
                                        void* user_data);
 
 /*
- * AOUT pre-fills sample_rate, samples_per_channel, data, and data_bytes before
+ * AOUT pre-fills format, samples_per_channel, data, and data_bytes before
  * calling render. The callback contract expects a non-negative rendered PCM
  * byte count. Backends defensively clamp negative returns to zero and zero-fill
  * any remaining buffer space.

package/vendor/runtime/macos-arm64/include/tirtc/audio_processing.h

@@ -2,6 +2,7 @@
 #define TIRTC_AUDIO_PROCESSING_H_
 
 #include "tirtc/audio_frame.h"
+#include "tirtc/audio_sample_rate.h"
 #include "tirtc/error.h"
 
 #ifdef __cplusplus

package/vendor/runtime/macos-arm64/include/tirtc/av.h

@@ -51,6 +51,7 @@ typedef struct TirtcVideoOutputMetricsSnapshot {
   double decoded_fps;
   double render_fps;
   uint32_t rate_window_duration_ms;
+  TirtcOutputLocalLatencyMetrics local_latency;
 } TirtcVideoOutputMetricsSnapshot;
 
 /**

package/vendor/runtime/macos-arm64/include/tirtc/media_codec.h

@@ -8,6 +8,7 @@ extern "C" {
 typedef enum TirtcMediaCodec {
   TIRTC_MEDIA_CODEC_NONE = 0,
   TIRTC_MEDIA_CODEC_AUDIO_G711A = 1,
+  TIRTC_MEDIA_CODEC_AUDIO_AAC = 2,
   /* Reserved shared identifiers; concrete video implementations live outside media-contract. */
   TIRTC_MEDIA_CODEC_VIDEO_H264 = 65,
   TIRTC_MEDIA_CODEC_VIDEO_H265 = 66,

package/vendor/runtime/macos-arm64/include/tirtc/media_downlink.h

@@ -66,6 +66,22 @@ typedef struct TirtcMediaDownlinkPendingMetricsSnapshot {
   uint64_t decoded_pending_duration_ms;
 } TirtcMediaDownlinkPendingMetricsSnapshot;
 
+typedef struct TirtcMediaDownlinkLatencyStageMetricsSnapshot {
+  uint64_t latest_ms;
+  uint64_t average_ms;
+  uint64_t peak_ms;
+  uint64_t sample_count;
+  uint64_t unavailable_count;
+} TirtcMediaDownlinkLatencyStageMetricsSnapshot;
+
+typedef struct TirtcMediaDownlinkLocalLatencyMetricsSnapshot {
+  uint32_t window_duration_ms;
+  TirtcMediaDownlinkLatencyStageMetricsSnapshot total;
+  TirtcMediaDownlinkLatencyStageMetricsSnapshot buffer;
+  TirtcMediaDownlinkLatencyStageMetricsSnapshot decode_or_ready;
+  TirtcMediaDownlinkLatencyStageMetricsSnapshot output;
+} TirtcMediaDownlinkLocalLatencyMetricsSnapshot;
+
 typedef struct TirtcMediaDownlinkAudioMetricsSnapshot {
   TirtcMediaDownlinkStutterMetricsSnapshot stutter;
   TirtcMediaDownlinkPendingMetricsSnapshot pending;
@@ -73,6 +89,7 @@ typedef struct TirtcMediaDownlinkAudioMetricsSnapshot {
   double input_packet_rate;
   double render_callback_rate;
   uint32_t rate_window_duration_ms;
+  TirtcMediaDownlinkLocalLatencyMetricsSnapshot local_latency;
 } TirtcMediaDownlinkAudioMetricsSnapshot;
 
 typedef struct TirtcMediaDownlinkVideoStartupMetricsSnapshot {
@@ -93,6 +110,7 @@ typedef struct TirtcMediaDownlinkVideoMetricsSnapshot {
   double decoded_fps;
   double render_fps;
   uint32_t rate_window_duration_ms;
+  TirtcMediaDownlinkLocalLatencyMetricsSnapshot local_latency;
 } TirtcMediaDownlinkVideoMetricsSnapshot;
 
 /**
@@ -103,6 +121,7 @@ typedef struct TirtcMediaDownlinkVideoMetricsSnapshot {
  */
 typedef struct TirtcMediaDownlinkAudioDebugSnapshot {
   TirtcMediaCodec codec;
+  TirtcAudioFormat format;
 } TirtcMediaDownlinkAudioDebugSnapshot;
 
 /**

package/vendor/runtime/macos-arm64/include/tirtc/media_uplink.h

@@ -29,7 +29,8 @@ typedef enum TirtcMediaUplinkState {
 } TirtcMediaUplinkState;
 
 typedef struct TirtcMediaUplinkAudioOptions {
-  TirtcAudioSampleRate sample_rate;
+  TirtcMediaCodec codec;
+  TirtcAudioFormat format;
   TirtcAudioAecMode aec_mode;
   TirtcAudioAgcLevel agc_level;
   TirtcAudioAnsLevel ans_level;

package/vendor/runtime/macos-arm64/manifest.txt

@@ -1,26 +1,27 @@
 platform=macos-arm64
 profile=full
-staged_at_utc=2026-05-09T08:57:10Z
+staged_at_utc=2026-05-13T16:53:41Z
 source_sdk=/Users/allenfeng/Development/Repositories/tirtc-nexus/tirtc-matrix/.build/sdk/macos-arm64
 
-92ac54ea714e5a7da536d3d63b955a9e30faf9560ebe670034a79f3ea69c21cb  include/tirtc/audio.h
-6d972ccfe150a3b4f2d7f18fa92b3ade9210c1c8bb754d061ac6b7997b59e2cb  include/tirtc/audio_codec.h
-7bacbdb2d8bb10d6444036a8fef42f2a8e3ea34dfc38e165ee678d61f189db41  include/tirtc/audio_frame.h
-fb8ea35263167c33487bd3aa503c26ed2dfbf97877f7bda92c10065cd5c97c13  include/tirtc/audio_io.h
+04c3139f26d53aa2fb0a586a21ffb481a001c01e0412cd3414a952c9a9d336ba  include/tirtc/audio.h
+ce7ccffa054250a2dab0e2128e4701a66c7168e50e8c470dcff3bf609ff0959c  include/tirtc/audio_codec.h
+0b463d8adfcf45acb3f95fe3e2d94b37a26aaf37d293eb2b18274ddcb161f767  include/tirtc/audio_format.h
+7cac5b8c0cef28526ee00800fa8b0d1e60897ab971a0801899cb0333c9189fde  include/tirtc/audio_frame.h
+6daadba731521c01a3cf11c5d8b2ef797238737c939e2da3045fc83ad89af04a  include/tirtc/audio_io.h
 c2e1f31dcc75be461c577d18b1cebe32774f212d51cb4dd2a5b5a9bfe62b693e  include/tirtc/audio_io_android.h
 878bb18220ac62304ec95807f31e83e050c8dac14811f873454f9b045028694a  include/tirtc/audio_io_apple.h
 0569eb1d043ee7fa8d45ca43143eca1a22e4c4bf1507ca7e045e010a3ecbc288  include/tirtc/audio_io_harmony.h
 51cbc911fe9f9834046f0e0a1a7cdd814a8e194a615894a8b4d11f9e5f095610  include/tirtc/audio_io_windows.h
-21f60729117260a44af22c1af986ef17d22673b102b7b7a035f492d0665cce16  include/tirtc/audio_processing.h
+5acfd19fdcd8c9d96e4aee7c6f1d33a9211c0a5b0c0ab56d96e1b4dee48f3aa9  include/tirtc/audio_processing.h
 0ca7c3c630b1242f51a0fd8154097c0a332b4c816a5707090e4381719852998c  include/tirtc/audio_sample_rate.h
-f40ce007241e4aa6ef2e2bcd589060636ef0e9859a9102c8b955d348f4f7eabc  include/tirtc/av.h
+07ca1b210a6e334c1474f25446477681472154568ecf74557cb3dc9b3a730662  include/tirtc/av.h
 b3e110ce04057d3be696b120d1abeaf6141711a1c63ca7f9bdf65ed5ec6e16d7  include/tirtc/error.h
 ae805545a9515edc9b94262e72ad2c7b7d649288166f4daeb450d8a55e82ae0b  include/tirtc/foundation/build_info.h
 7cf8b372a3d48d4de4a65a04c7f102281a7b42cebb9ec247853d3c53afb63b6a  include/tirtc/http.h
 70bbf93b84d9d1a85f376d9986de570c1f658319e1e5ab6d621f7a4d41033f5c  include/tirtc/logging.h
-7e2f7f81afe37e22f5414d9db32438edd948ff775230c3a0a17ff9ac1a0b560e  include/tirtc/media_codec.h
-43c0ac9173b6259a1b1379eac780f4892a0a72572801ac9eb8516c0bdb4913c3  include/tirtc/media_downlink.h
-e64191a39004165cdcd11b2ca824af34d4bfe02f7ec3c3101fa9291f56488c30  include/tirtc/media_uplink.h
+1419ac3fb5237d1d820bd30a6845b98aab3553463eb64e73623b64c663665f16  include/tirtc/media_codec.h
+df11627da9bfbb87cb32582701f638d2a0ff5b6b77b49a3a3bab7509c62f0613  include/tirtc/media_downlink.h
+9376af63fde915753146969291f0fefe2da6df28f7aa6310f7787048165150dc  include/tirtc/media_uplink.h
 57be6941026e7dd173e8779b2bb868532d65e41f7917aad476a31ac25c1cd6c3  include/tirtc/transport.h
 fba7954919606e507b6a074440440903665217959699da89ca9dd206772c0ef0  include/tirtc/trp.h
 92b3db6e2cb02265ad5db01881edce9d4e095b1dadf449dce5d8bb5362957e33  include/tirtc/video_codec.h
@@ -32,16 +33,16 @@ cae0bbeb884e5466a56da15182c78cc22baab6c743f349a58d3595f623333585  include/tirtc/
 65b5d24fe3a6bc2a299a2220148a153ba4eb0955a49c1c3baabf1fdfed8aabdd  include/tirtc/video_io_windows.h
 8cd6b66bea14890a665cc317f8572429b2c3e4463773f8b77c1e4dc30a4a8747  include/tirtc/video_processing.h
 5052a464a34508f117fec46aa6b836871014229d4290d7d8caac862e0165ae8f  lib/libTGTRP.a
-8db86d6714264047e8fd4086ddd7315722d675749719e6175f89eb5a636b48a1  lib/libTiRTC.a
+41d64a3a9dd5cd4f51d45be501e8df2161a59847bb599b218ac55194673189f9  lib/libTiRTC.a
 b39daee6a3d39bf0ca20c45084601133c4198de8dca848dcff6dd9c70ae99016  lib/libcrypto.a
 c052857ef315e3d61db9c862cad10709a3a6b2487dc41799cbe4d74a805de875  lib/libcrypto.dylib
-e2cc6ebe4b3efd602dd454c32cf5a56f7cecddff52ba51a09307434a102a3134  lib/libmatrix_runtime_audio.a
-86952647a2ee67457e5e0129270bbcc130bc1c3055674bd1256996e63d1d396d  lib/libmatrix_runtime_facade.a
-88c8067852e3a334f6d7b741d36ffe4d5dc5a639b50a5566178485f2e3a62d51  lib/libmatrix_runtime_foundation_http.a
-5928ccfe2717594f66aba6b43d3b3d394634d2d742d5a79ccd86d17ab56adf6c  lib/libmatrix_runtime_foundation_logging.a
-a1a7c04dff1601fee215bf71b99a1b2725b74171b7182ab807b4f40ba96a5554  lib/libmatrix_runtime_media.a
-8adcf9d35a21d86392336389e361d6c75554ba1cdd6afde74e9e748adff34b77  lib/libmatrix_runtime_transport.a
-9e8db21a0c02d407799a9acc80042af2de38bc2b9509914df6d0bd63cae1a945  lib/libmatrix_runtime_video.a
+3316e224dda38d03604c68a3904efddb13909dfc50d682c24a3252679646ffc9  lib/libmatrix_runtime_audio.a
+50b5deab7ecfeb7cbf5fe9ab6f91a6d163ebcf518c6e3beb0096a33db1833640  lib/libmatrix_runtime_facade.a
+a7a3f8e4adac588583db2dc6179aec86ae73e93cb65a85d14933dd7b2724fb3f  lib/libmatrix_runtime_foundation_http.a
+e602bd53bebb54f5e612fccade3add7abd418e235c41974b532739f3218c5fe1  lib/libmatrix_runtime_foundation_logging.a
+74e1b90aedaf84c7679c65278804a27fa5fd8d8101d05dfed9782728a15719cd  lib/libmatrix_runtime_media.a
+16a2dc91c75f99fa9cef357cefe89cf0fdfef682de0524d18b08e6d8e86875cf  lib/libmatrix_runtime_transport.a
+c236e578105e2bf1551902f763426ec53a8ae1e40d44833e2ba7b16afc64d00b  lib/libmatrix_runtime_video.a
 c11c65d373a127028350c41fa58cd2d1223f2b5d70a84e13b115d90daaba25ca  lib/libssl.a
 ef1c1104bbdd2528ed7b958fb7252bd6249875f92300b0c9577d6c4bd6c0d88a  lib/libssl.dylib
 e14e846e43d64e240fa0e5745bf4e702b79d0f2442e7f768beb990610735c71b  lib/libtgrtc.dylib

package/vendor/runtime/script/prepare_runtime_media_dataset.sh

@@ -133,19 +133,19 @@ require_cmd "$FFPROBE_BIN"
 SOURCE_ABS="$(cd "$(dirname "$SOURCE")" && pwd)/$(basename "$SOURCE")"
 SOURCE_FILE_NAME="$(basename "$SOURCE_ABS")"
 
-readonly k_schema_version=2
-readonly k_tool_version="runtime_local_assets_prepare_v5"
+readonly k_schema_version=3
+readonly k_tool_version="runtime_local_assets_prepare_v6"
 readonly k_video_max_width=1280
 readonly k_video_max_height=720
 readonly k_video_fps=15
 readonly k_audio_sample_rate_hz=8000
 readonly k_audio_sample_rate_16k_hz=16000
-readonly k_audio_channels=1
 readonly k_audio_packet_duration_ms=40
+readonly k_audio_bits_per_sample=16
+readonly k_aac_packet_samples_per_channel=1024
 readonly k_mjpeg_packet_duration_ms=80
 readonly k_mjpeg_extract_interval_seconds=10
 readonly k_loop_policy="continuous_pts"
-readonly k_audio_codec="g711a"
 readonly k_output_audio_format="pcm_s16"
 readonly k_output_video_format="rgba8888"
 
@@ -253,10 +253,6 @@ VIDEO_H265_PATH="$ASSETS_DIR/video/video.h265"
 VIDEO_H265_PACKET_INDEX_PATH="$ASSETS_DIR/video/video_h265_packets.csv"
 VIDEO_MJPEG_PATH="$ASSETS_DIR/video/video.mjpeg"
 VIDEO_MJPEG_PACKET_INDEX_PATH="$ASSETS_DIR/video/video_mjpeg_packets.csv"
-AUDIO_PATH="$ASSETS_DIR/audio/audio.g711a"
-AUDIO_PACKET_INDEX_PATH="$ASSETS_DIR/audio/audio_packets.csv"
-AUDIO_16K_PATH="$ASSETS_DIR/audio/audio_16k.g711a"
-AUDIO_16K_PACKET_INDEX_PATH="$ASSETS_DIR/audio/audio_16k_packets.csv"
 
 video_encode_args=(
   -vf "fps=${k_video_fps},scale=${TARGET_VIDEO_WIDTH}:${TARGET_VIDEO_HEIGHT}:flags=lanczos"
@@ -335,26 +331,6 @@ if ! find "$mjpeg_frame_dir" -type f -name 'frame_*.jpg' | grep -q .; then
     "$mjpeg_frame_dir/frame_000001.jpg"
 fi
 
-progress "encoding 8k g711a audio track"
-"$FFMPEG_BIN" -hide_banner -loglevel error -y \
-  -i "$SOURCE_ABS" \
-  -vn \
-  -ac ${k_audio_channels} \
-  -ar ${k_audio_sample_rate_hz} \
-  -c:a pcm_alaw \
-  -f alaw \
-  "$AUDIO_PATH"
-
-progress "encoding 16k g711a audio track"
- "$FFMPEG_BIN" -hide_banner -loglevel error -y \
-  -i "$SOURCE_ABS" \
-  -vn \
-  -ac ${k_audio_channels} \
-  -ar ${k_audio_sample_rate_16k_hz} \
-  -c:a pcm_alaw \
-  -f alaw \
-  "$AUDIO_16K_PATH"
-
 progress "indexing h264 video packets"
 VIDEO_PACKET_INDEX_PATH="$VIDEO_H264_PACKET_INDEX_PATH" \
 VIDEO_PATH="$VIDEO_H264_PATH" \
@@ -607,16 +583,35 @@ for (let index = 0; index < packetCount; index += 1) {
 fs.writeFileSync(indexPath, `${lines.join('\n')}\n`, 'utf8');
 NODE
 
-progress "indexing 8k g711a audio packets"
-AUDIO_BYTES="$(wc -c < "$AUDIO_PATH" | tr -d '[:space:]')"
-AUDIO_PACKET_BYTES=$((k_audio_sample_rate_hz * k_audio_packet_duration_ms / 1000))
-[[ "$AUDIO_PACKET_BYTES" -gt 0 ]] || fail "invalid audio packet bytes"
-AUDIO_REMAINDER=$((AUDIO_BYTES % AUDIO_PACKET_BYTES))
-if [[ "$AUDIO_REMAINDER" -ne 0 ]]; then
-  AUDIO_ALIGNED_BYTES=$((AUDIO_BYTES - AUDIO_REMAINDER))
-  [[ "$AUDIO_ALIGNED_BYTES" -gt 0 ]] || fail "audio output is too short for one full packet"
-  echo "[prepare_runtime_media_dataset] trim trailing audio bytes to packet boundary: removed=${AUDIO_REMAINDER}" >&2
-  AUDIO_PATH="$AUDIO_PATH" AUDIO_ALIGNED_BYTES="$AUDIO_ALIGNED_BYTES" node <<'NODE'
+encode_g711a_track() {
+  local sample_rate="$1"
+  local channels="$2"
+  local key="$3"
+  local output_path="$ASSETS_DIR/audio/${key}.g711a"
+  local index_path="$ASSETS_DIR/audio/${key}.csv"
+
+  progress "encoding ${key} audio track"
+  "$FFMPEG_BIN" -hide_banner -loglevel error -y \
+    -i "$SOURCE_ABS" \
+    -vn \
+    -ac "$channels" \
+    -ar "$sample_rate" \
+    -c:a pcm_alaw \
+    -f alaw \
+    "$output_path"
+
+  progress "indexing ${key} audio packets"
+  local audio_bytes
+  audio_bytes="$(wc -c < "$output_path" | tr -d '[:space:]')"
+  local samples_per_packet=$((sample_rate * k_audio_packet_duration_ms / 1000))
+  local packet_bytes=$((samples_per_packet * channels))
+  [[ "$packet_bytes" -gt 0 ]] || fail "invalid audio packet bytes"
+  local remainder=$((audio_bytes % packet_bytes))
+  if [[ "$remainder" -ne 0 ]]; then
+    local aligned_bytes=$((audio_bytes - remainder))
+    [[ "$aligned_bytes" -gt 0 ]] || fail "${key} audio output is too short for one full packet"
+    echo "[prepare_runtime_media_dataset] trim trailing ${key} bytes to packet boundary: removed=${remainder}" >&2
+    AUDIO_PATH="$output_path" AUDIO_ALIGNED_BYTES="$aligned_bytes" node <<'NODE'
 const fs = require('fs');
 const audioPath = process.env.AUDIO_PATH;
 const alignedBytes = Number(process.env.AUDIO_ALIGNED_BYTES || '0');
@@ -625,51 +620,93 @@ if (!audioPath || !Number.isFinite(alignedBytes) || alignedBytes <= 0) {
 }
 fs.truncateSync(audioPath, alignedBytes);
 NODE
-  AUDIO_BYTES="$AUDIO_ALIGNED_BYTES"
-fi
-AUDIO_PACKET_COUNT=$((AUDIO_BYTES / AUDIO_PACKET_BYTES))
-[[ "$AUDIO_PACKET_COUNT" -gt 0 ]] || fail "generated audio packets are empty"
-
-{
-  echo "pts_us,offset,size"
-  for ((i = 0; i < AUDIO_PACKET_COUNT; i += 1)); do
-    pts_us=$((i * k_audio_packet_duration_ms * 1000))
-    offset=$((i * AUDIO_PACKET_BYTES))
-    echo "${pts_us},${offset},${AUDIO_PACKET_BYTES}"
-  done
-} > "$AUDIO_PACKET_INDEX_PATH"
-
-progress "indexing 16k g711a audio packets"
-AUDIO_16K_BYTES="$(wc -c < "$AUDIO_16K_PATH" | tr -d '[:space:]')"
-AUDIO_16K_PACKET_BYTES=$((k_audio_sample_rate_16k_hz * k_audio_packet_duration_ms / 1000))
-[[ "$AUDIO_16K_PACKET_BYTES" -gt 0 ]] || fail "invalid 16k audio packet bytes"
-AUDIO_16K_REMAINDER=$((AUDIO_16K_BYTES % AUDIO_16K_PACKET_BYTES))
-if [[ "$AUDIO_16K_REMAINDER" -ne 0 ]]; then
-  AUDIO_16K_ALIGNED_BYTES=$((AUDIO_16K_BYTES - AUDIO_16K_REMAINDER))
-  [[ "$AUDIO_16K_ALIGNED_BYTES" -gt 0 ]] || fail "16k audio output is too short for one full packet"
-  echo "[prepare_runtime_media_dataset] trim trailing 16k audio bytes to packet boundary: removed=${AUDIO_16K_REMAINDER}" >&2
-  AUDIO_PATH="$AUDIO_16K_PATH" AUDIO_ALIGNED_BYTES="$AUDIO_16K_ALIGNED_BYTES" node <<'NODE'
+    audio_bytes="$aligned_bytes"
+  fi
+  local packet_count=$((audio_bytes / packet_bytes))
+  [[ "$packet_count" -gt 0 ]] || fail "generated ${key} audio packets are empty"
+
+  {
+    echo "pts_us,offset,size,samples_per_channel"
+    for ((i = 0; i < packet_count; i += 1)); do
+      pts_us=$((i * k_audio_packet_duration_ms * 1000))
+      offset=$((i * packet_bytes))
+      echo "${pts_us},${offset},${packet_bytes},${samples_per_packet}"
+    done
+  } > "$index_path"
+}
+
+encode_aac_track() {
+  local sample_rate="$1"
+  local channels="$2"
+  local key="$3"
+  local adts_path="$ASSETS_DIR/audio/${key}.adts"
+  local output_path="$ASSETS_DIR/audio/${key}.aac"
+  local index_path="$ASSETS_DIR/audio/${key}.csv"
+
+  progress "encoding ${key} audio track"
+  "$FFMPEG_BIN" -hide_banner -loglevel error -y \
+    -i "$SOURCE_ABS" \
+    -vn \
+    -ac "$channels" \
+    -ar "$sample_rate" \
+    -c:a aac \
+    -profile:a aac_low \
+    -f adts \
+    "$adts_path"
+
+  progress "indexing ${key} audio packets"
+  ADTS_PATH="$adts_path" OUTPUT_PATH="$output_path" INDEX_PATH="$index_path" SAMPLE_RATE="$sample_rate" AAC_PACKET_SAMPLES_PER_CHANNEL="$k_aac_packet_samples_per_channel" node <<'NODE'
 const fs = require('fs');
-const audioPath = process.env.AUDIO_PATH;
-const alignedBytes = Number(process.env.AUDIO_ALIGNED_BYTES || '0');
-if (!audioPath || !Number.isFinite(alignedBytes) || alignedBytes <= 0) {
-  throw new Error('audio trim env is incomplete');
+const adtsPath = process.env.ADTS_PATH;
+const outputPath = process.env.OUTPUT_PATH;
+const indexPath = process.env.INDEX_PATH;
+const sampleRate = Number(process.env.SAMPLE_RATE || '0');
+if (!adtsPath || !outputPath || !indexPath || !Number.isFinite(sampleRate) || sampleRate <= 0) {
+  throw new Error('aac packet index env is incomplete');
+}
+const data = fs.readFileSync(adtsPath);
+const chunks = [];
+const rows = ['pts_us,offset,size,samples_per_channel'];
+let cursor = 0;
+let outputOffset = 0;
+let packetIndex = 0;
+const samplesPerChannel = Number(process.env.AAC_PACKET_SAMPLES_PER_CHANNEL || '1024');
+if (!Number.isFinite(samplesPerChannel) || samplesPerChannel <= 0) {
+  throw new Error('aac packet samples env is invalid');
+}
+while (cursor + 7 <= data.length) {
+  if (data[cursor] !== 0xff || (data[cursor + 1] & 0xf0) !== 0xf0) {
+    throw new Error(`aac adts sync missing at offset ${cursor}`);
+  }
+  const protectionAbsent = data[cursor + 1] & 0x01;
+  const headerBytes = protectionAbsent ? 7 : 9;
+  const frameLength = ((data[cursor + 3] & 0x03) << 11) | (data[cursor + 4] << 3) | ((data[cursor + 5] & 0xe0) >> 5);
+  if (frameLength <= headerBytes || cursor + frameLength > data.length) {
+    throw new Error(`aac adts invalid frame length at offset ${cursor}`);
+  }
+  const payload = data.subarray(cursor + headerBytes, cursor + frameLength);
+  const ptsUs = Math.round(packetIndex * samplesPerChannel * 1_000_000 / sampleRate);
+  rows.push(`${ptsUs},${outputOffset},${payload.length},${samplesPerChannel}`);
+  chunks.push(payload);
+  outputOffset += payload.length;
+  packetIndex += 1;
+  cursor += frameLength;
 }
-fs.truncateSync(audioPath, alignedBytes);
+if (packetIndex === 0 || outputOffset === 0) {
+  throw new Error('aac packet index is empty');
+}
+fs.writeFileSync(outputPath, Buffer.concat(chunks));
+fs.writeFileSync(indexPath, `${rows.join('\n')}\n`, 'utf8');
+fs.rmSync(adtsPath, {force: true});
 NODE
-  AUDIO_16K_BYTES="$AUDIO_16K_ALIGNED_BYTES"
-fi
-AUDIO_16K_PACKET_COUNT=$((AUDIO_16K_BYTES / AUDIO_16K_PACKET_BYTES))
-[[ "$AUDIO_16K_PACKET_COUNT" -gt 0 ]] || fail "generated 16k audio packets are empty"
+}
 
-{
-  echo "pts_us,offset,size"
-  for ((i = 0; i < AUDIO_16K_PACKET_COUNT; i += 1)); do
-    pts_us=$((i * k_audio_packet_duration_ms * 1000))
-    offset=$((i * AUDIO_16K_PACKET_BYTES))
-    echo "${pts_us},${offset},${AUDIO_16K_PACKET_BYTES}"
+for sample_rate in "$k_audio_sample_rate_hz" "$k_audio_sample_rate_16k_hz"; do
+  for channels in 1 2; do
+    encode_g711a_track "$sample_rate" "$channels" "g711a_${sample_rate}_${channels}ch_s16"
+    encode_aac_track "$sample_rate" "$channels" "aac_${sample_rate}_${channels}ch_s16"
   done
-} > "$AUDIO_16K_PACKET_INDEX_PATH"
+done
 
 progress "writing prepared media manifest"
 cat > "$MANIFEST_PATH" <<JSON
@@ -690,12 +727,14 @@ cat > "$MANIFEST_PATH" <<JSON
     "bitrate_kbps": ${TARGET_VIDEO_BITRATE_KBPS}
   },
   "audio": {
-    "path": "audio/audio.g711a",
-    "packet_index_path": "audio/audio_packets.csv",
-    "codec": "${k_audio_codec}",
+    "path": "audio/g711a_8000_1ch_s16.g711a",
+    "packet_index_path": "audio/g711a_8000_1ch_s16.csv",
+    "codec": "g711a",
     "sample_rate_hz": ${k_audio_sample_rate_hz},
-    "channels": ${k_audio_channels},
-    "packet_duration_ms": ${k_audio_packet_duration_ms}
+    "channels": 1,
+    "bits_per_sample": ${k_audio_bits_per_sample},
+    "packet_duration_ms": ${k_audio_packet_duration_ms},
+    "packet_samples_per_channel": $((k_audio_sample_rate_hz * k_audio_packet_duration_ms / 1000))
   },
   "output": {
     "audio_format": "${k_output_audio_format}",
@@ -731,21 +770,85 @@ cat > "$MANIFEST_PATH" <<JSON
     }
   },
   "audio_tracks": {
-    "g711a_8k": {
-      "path": "audio/audio.g711a",
-      "packet_index_path": "audio/audio_packets.csv",
-      "codec": "${k_audio_codec}",
+    "g711a_8000_1ch_s16": {
+      "path": "audio/g711a_8000_1ch_s16.g711a",
+      "packet_index_path": "audio/g711a_8000_1ch_s16.csv",
+      "codec": "g711a",
+      "sample_rate_hz": ${k_audio_sample_rate_hz},
+      "channels": 1,
+      "bits_per_sample": ${k_audio_bits_per_sample},
+      "packet_duration_ms": ${k_audio_packet_duration_ms},
+      "packet_samples_per_channel": $((k_audio_sample_rate_hz * k_audio_packet_duration_ms / 1000))
+    },
+    "g711a_8000_2ch_s16": {
+      "path": "audio/g711a_8000_2ch_s16.g711a",
+      "packet_index_path": "audio/g711a_8000_2ch_s16.csv",
+      "codec": "g711a",
       "sample_rate_hz": ${k_audio_sample_rate_hz},
-      "channels": ${k_audio_channels},
-      "packet_duration_ms": ${k_audio_packet_duration_ms}
+      "channels": 2,
+      "bits_per_sample": ${k_audio_bits_per_sample},
+      "packet_duration_ms": ${k_audio_packet_duration_ms},
+      "packet_samples_per_channel": $((k_audio_sample_rate_hz * k_audio_packet_duration_ms / 1000))
+    },
+    "g711a_16000_1ch_s16": {
+      "path": "audio/g711a_16000_1ch_s16.g711a",
+      "packet_index_path": "audio/g711a_16000_1ch_s16.csv",
+      "codec": "g711a",
+      "sample_rate_hz": ${k_audio_sample_rate_16k_hz},
+      "channels": 1,
+      "bits_per_sample": ${k_audio_bits_per_sample},
+      "packet_duration_ms": ${k_audio_packet_duration_ms},
+      "packet_samples_per_channel": $((k_audio_sample_rate_16k_hz * k_audio_packet_duration_ms / 1000))
+    },
+    "g711a_16000_2ch_s16": {
+      "path": "audio/g711a_16000_2ch_s16.g711a",
+      "packet_index_path": "audio/g711a_16000_2ch_s16.csv",
+      "codec": "g711a",
+      "sample_rate_hz": ${k_audio_sample_rate_16k_hz},
+      "channels": 2,
+      "bits_per_sample": ${k_audio_bits_per_sample},
+      "packet_duration_ms": ${k_audio_packet_duration_ms},
+      "packet_samples_per_channel": $((k_audio_sample_rate_16k_hz * k_audio_packet_duration_ms / 1000))
+    },
+    "aac_8000_1ch_s16": {
+      "path": "audio/aac_8000_1ch_s16.aac",
+      "packet_index_path": "audio/aac_8000_1ch_s16.csv",
+      "codec": "aac",
+      "sample_rate_hz": ${k_audio_sample_rate_hz},
+      "channels": 1,
+      "bits_per_sample": ${k_audio_bits_per_sample},
+      "packet_duration_ms": $((k_aac_packet_samples_per_channel * 1000 / k_audio_sample_rate_hz)),
+      "packet_samples_per_channel": ${k_aac_packet_samples_per_channel}
+    },
+    "aac_8000_2ch_s16": {
+      "path": "audio/aac_8000_2ch_s16.aac",
+      "packet_index_path": "audio/aac_8000_2ch_s16.csv",
+      "codec": "aac",
+      "sample_rate_hz": ${k_audio_sample_rate_hz},
+      "channels": 2,
+      "bits_per_sample": ${k_audio_bits_per_sample},
+      "packet_duration_ms": $((k_aac_packet_samples_per_channel * 1000 / k_audio_sample_rate_hz)),
+      "packet_samples_per_channel": ${k_aac_packet_samples_per_channel}
+    },
+    "aac_16000_1ch_s16": {
+      "path": "audio/aac_16000_1ch_s16.aac",
+      "packet_index_path": "audio/aac_16000_1ch_s16.csv",
+      "codec": "aac",
+      "sample_rate_hz": ${k_audio_sample_rate_16k_hz},
+      "channels": 1,
+      "bits_per_sample": ${k_audio_bits_per_sample},
+      "packet_duration_ms": $((k_aac_packet_samples_per_channel * 1000 / k_audio_sample_rate_16k_hz)),
+      "packet_samples_per_channel": ${k_aac_packet_samples_per_channel}
     },
-    "g711a_16k": {
-      "path": "audio/audio_16k.g711a",
-      "packet_index_path": "audio/audio_16k_packets.csv",
-      "codec": "${k_audio_codec}",
+    "aac_16000_2ch_s16": {
+      "path": "audio/aac_16000_2ch_s16.aac",
+      "packet_index_path": "audio/aac_16000_2ch_s16.csv",
+      "codec": "aac",
       "sample_rate_hz": ${k_audio_sample_rate_16k_hz},
-      "channels": ${k_audio_channels},
-      "packet_duration_ms": ${k_audio_packet_duration_ms}
+      "channels": 2,
+      "bits_per_sample": ${k_audio_bits_per_sample},
+      "packet_duration_ms": $((k_aac_packet_samples_per_channel * 1000 / k_audio_sample_rate_16k_hz)),
+      "packet_samples_per_channel": ${k_aac_packet_samples_per_channel}
     }
   },
   "loop_policy": "${k_loop_policy}"