Voice Activity Detection

Detects the speech regions of an audio or video file (JSON: base64 or file identifier).#

Runs the built-in Silero voice-activity-detection model over the audio and returns the time ranges that contain speech, with the total speech time and the audio duration. No transcription model is involved: the pass is CPU-only and needs nothing downloaded or loaded, so it is fast even on servers without a speech model. Typical uses: segmenting a recording before transcription, skipping silence in long audio, or measuring how much of a file is actually spoken. The optional 'vad' object tunes detection (threshold, minimum durations, padding).

Accepts an audio or video file as base64 (input_format 'Base64EncodedFile') or as the identifier of a previously uploaded file (input_format 'FileIdentifier'). Decoding shares the transcription pipeline: the formats the server can decode are reported by GET /lmkit/v1/audio-transcription/capabilities, and a video's audio track is extracted. A video that declares no audio track answers 422 Unprocessable Content. If processing exceeds the configured timeout, returns 202 Accepted with a job_id; poll GET /lmkit/v1/jobs/ for status and results.

Large files: the base64 body is bounded by the configured "Max upload size" (default 100 MB), but base64 inflates a file by ~33% and is buffered whole in memory. For large or long recordings prefer either (a) POST the raw file to /lmkit/v1/voice-activity-detection/file (multipart, streamed to disk), or (b) POST it once to /lmkit/v1/files/upload and reference the returned fileId here with input_format 'FileIdentifier'. An oversized body is rejected with 413 Payload Too Large.

Request body

application/json · VoiceActivityDetectionRequest

Responses

curl -X POST "$LMKIT_ONE_URL/lmkit/v1/voice-activity-detection" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "input": "",
  "input_format": "Base64EncodedFile",
  "vad": {
    "energy_threshold": "string",
    "min_speech_ms": "string",
    "min_silence_ms": "string",
    "max_speech_s": "string"
  }
}'

Detects the speech regions of an uploaded audio or video file (multipart/form-data).#

Accepts an audio or video file as a raw multipart/form-data upload (form field 'file'). This is the recommended path for large or long recordings: the file is streamed to disk rather than buffered whole in memory, and no base64 inflation applies. The upload is bounded by the configured "Max upload size" (default 100 MB); an oversized file is rejected with 413 Payload Too Large. Detection tuning is available on the JSON endpoint's 'vad' field. Behaves identically to the JSON endpoint otherwise: returns the speech segments with the total speech time, or 202 Accepted with a job_id when processing exceeds the configured timeout (poll GET /lmkit/v1/jobs/).

Request body

application/x-www-form-urlencoded · object

Responses

curl -X POST "$LMKIT_ONE_URL/lmkit/v1/voice-activity-detection/file" \
  -H "Authorization: Bearer $LMKIT_API_KEY" \
  -H "Content-Type: application/x-www-form-urlencoded"