Set the VAD threshold for speech-to-text

vad_threshold (0.0–1.0) gates which audio segments count as speech. Segments below the threshold are skipped for transcription. 0 disables the gate. Lower values pick up quieter or noisier speech (narrowband telephony) and can produce spurious text for background noise.

Defaults differ by surface: batch REST defaults to 0.5; streaming WebSocket defaults to 0.08.

Batch REST

POST https://api.x.ai/v1/stt. Put option fields before file in the multipart body. Fields after file may be ignored.

Official skeleton, with vad_threshold added before file:

curl -X POST https://api.x.ai/v1/stt \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -F format=true \
  -F language=en \
  -F "keyterm=Understand The Universe" \
  -F vad_threshold=0.5 \
  -F file=@audio.mp3

Same order in requests: options in data=, file in files=.

import os
import requests

response = requests.post(
    "https://api.x.ai/v1/stt",
    headers={"Authorization": f"Bearer {os.environ['XAI_API_KEY']}"},
    files={"file": ("audio.mp3", open("audio.mp3", "rb"), "audio/mpeg")},
    data=[
        ("format", "true"),
        ("language", "en"),
        ("keyterm", "Understand The Universe"),
        ("vad_threshold", "0.5"),
    ],
)

Streaming WebSocket

Add &vad_threshold=0.08 (or another value in 0.0–1.0) to the documented URL pattern:

wss://api.x.ai/v1/stt?sample_rate=16000&encoding=pcm&interim_results=true&language=en&keyterm=Understand+The+Universe

On streaming, the VAD gate does not change endpointing or speech_final timing. Other knobs on the same page include smart_turn / smart_turn_timeout and endpointing.

Pitfalls

• Batch default is 0.5. Streaming default is 0.08. Do not mix them.
• 0 disables the gate on both surfaces.
• Batch: option fields must precede file, or later fields may be ignored.
• Console API credits are separate from SuperGrok's weekly pool on grok.com.