菜单
On-chip Voice Activity Detection (VAD) for TuyaOpen via the tkl_vad API: detect speech vs silence from a mic audio stream. VAD、语音活动检测、人声检测、静音检测、断句、tkl_vad、是否在说话。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | tuyaopen/onchip-vad |
| description | On-chip Voice Activity Detection (VAD) for TuyaOpen via the tkl_vad API: detect speech vs silence from a mic audio stream. VAD、语音活动检测、人声检测、静音检测、断句、tkl_vad、是否在说话。 |
| when_to_use | Use when the user wants to detect whether someone is speaking (speech vs silence) from the microphone stream — e.g. gate recording, end-pointing. Needs mic audio (see the audio peripheral skill). |
| id | onchip-vad |
| surfaces | ["embedded"] |
| tags | ["vad","voice","speech-detect","audio","on-chip","tkl_vad"] |
VAD analyses microphone audio. Set up audio input first (see the audio peripheral skill), then feed captured PCM frames to VAD and poll its status.
CONFIG_ENABLE_VAD. No app_default.config change.board_register_hardware(). Call tkl_vad_* directly. Single engine (no pins) —
record as onchip:vad in used-peripherals.json if wanted.Config fields come from your audio format. Below are the real fields used by the SDK.
#include "tal_api.h"
#include "tkl_vad.h"
void app_vad_init(uint32_t sample_rate, uint8_t channels)
{
TKL_VAD_CONFIG_T cfg = {0};
cfg.sample_rate = sample_rate; /* match the mic, e.g. 16000 */
cfg.channel_num = channels; /* e.g. 1 */
cfg.speech_min_ms = 300; /* min speech to assert SPEECH */
cfg.noise_min_ms = 800; /* min silence to drop back */
cfg.frame_duration_ms = 10; /* feed in 10 ms frames */
cfg.scale = 1.0f;
tkl_vad_init(&cfg);
tkl_vad_start();
}
/* For each captured PCM frame: feed it, then check status. */
void app_vad_on_pcm(uint8_t *pcm, uint32_t len)
{
tkl_vad_feed(pcm, len);
if (tkl_vad_get_status() == TKL_VAD_STATUS_SPEECH) {
/* speaking */
} else {
/* silence (TKL_VAD_STATUS_NONE) */
}
}
Stop/restart around a session with tkl_vad_stop() / tkl_vad_start().
| Function | Description |
|---|---|
tkl_vad_init(&cfg) | cfg.{sample_rate, channel_num, speech_min_ms, noise_min_ms, frame_duration_ms, scale} |
tkl_vad_start() / tkl_vad_stop() | Begin / end detection |
tkl_vad_feed(data, len) | Feed a PCM frame |
tkl_vad_get_status() | TKL_VAD_STATUS_SPEECH (1) / TKL_VAD_STATUS_NONE (0) |
tkl_vad_deinit() | Release the engine |
| Symptom | Likely cause | Fix |
|---|---|---|
| Always silence | sample_rate/channel_num mismatch with the mic | Match the audio input format |
| Flaps speech/silence | Thresholds too tight | Raise speech_min_ms / noise_min_ms |
| No status changes | Not feeding, or feeding wrong frame size | Feed frame_duration_ms-sized PCM each call |
SDK: src/ai_components/ai_audio/src/ai_audio_input.c
(tkl_vad_init with the fields above, tkl_vad_start/feed/get_status/stop).
End-to-end IoT product development orchestration for TuyaOpen projects. Guides from requirements gathering → Tuya Platform product/DP creation → complete embedded firmware generation. State-machine: detects project state and picks up from wherever development currently stands.
TDL audio codec usage for TuyaOpen: discover the audio device, capture microphone PCM frames via callback, play PCM back to the speaker, and set volume. 音频、麦克风、录音、播放、喇叭、扬声器、speaker、mic、声音、语音、codec。
TDL button input handling for TuyaOpen: create button, register event callbacks for single click, double click, long press, and multi-click. 按键、单击、双击、长按、多击、按键事件。
TDL camera control for TuyaOpen: find device, configure resolution/fps/format, open with frame callbacks for YUV422 and JPEG/H264 encoded output. 摄像头、相机、拍照、视频流、YUV、JPEG、H264、人脸识别、图像采集。
TDL display control for TuyaOpen: driver registration, open, brightness, and framebuffer flush via tdl_disp_* APIs. 显示屏、LCD、屏幕、显示、绘制、亮度、刷屏。
TDL thermal printer usage for TuyaOpen: discover the printer, print UTF-8 text (auto GBK conversion on ESC/POS), print 1-bit bitmaps, feed paper, and monitor paper/temperature status. 打印机、热敏打印、打印、小票、票据、出纸、ESC/POS、printer、DP48A。