speech-to-text

name: speech-to-text description: Transcribe audio to text using the slng.ai Voice AI API. Use when transcribing audio files, building live captions, processing meeting recordings, or working with Deepgram Nova, Sarvam Saaras, Soniox, or Reson8 STT via slng. license: MIT compatibility: Requires internet access and a slng.ai API key (VOICEAI_API_KEY).

slng Speech-to-Text

Transcribe audio (file or live stream) through slng's unified STT API. One protocol reaches every provider (Deepgram, Sarvam, Soniox, Reson8). File transcription uses HTTP; live audio uses WebSocket.

Setup: setup-api-key configures VOICEAI_API_KEY. Install paths are in references/installation.md.

Quick Start

CLI

# Transcribe a file
voiceai stt audio.wav --model slng/deepgram/nova:3-en

# Live microphone
voiceai stt --stream

Python

import os, requests

with open("audio.wav", "rb") as f:
    r = requests.post(
        "https://api.slng.ai/v1/stt/slng/deepgram/nova:3-en",
        headers={"Authorization": f"Bearer {os.environ['VOICEAI_API_KEY']}"},
        files={"audio": f},
    )

print(r.json()["transcript"])

TypeScript

import { readFileSync } from "fs";

const form = new FormData();
form.append("audio", new Blob([readFileSync("audio.wav")]), "audio.wav");

const r = await fetch("https://api.slng.ai/v1/stt/slng/deepgram/nova:3-en", {
  method: "POST",
  headers: { Authorization: `Bearer ${process.env.VOICEAI_API_KEY}` },
  body: form,
});
console.log((await r.json()).transcript);

cURL

curl https://api.slng.ai/v1/stt/slng/deepgram/nova:3-en \
  -H "Authorization: Bearer $VOICEAI_API_KEY" \
  -F "audio=@audio.wav"

Models

Unified endpoint shape: POST /v1/stt/{provider}/{model}:{variant} for files, WSS /v1/stt/{provider}/{model}:{variant} for streaming.

List current deployments:

voiceai models --stt

Streaming (WebSocket)

For live mic input or low-latency captions, use the WebSocket endpoint. See references/streaming.md for the full protocol and examples in all 4 paths.

Quick CLI:

voiceai stt --stream                                  # mic
arecord -f S16_LE -r 16000 -c 1 | voiceai stt --stream --source stdin

File transcription

For pre-recorded audio, the HTTP endpoint is simplest. Supports common formats (WAV, MP3, FLAC, OGG, WebM, M4A). See references/file-transcription.md.

Response shape

File transcription returns JSON:

{
  "transcript": "Hello from slng.",
  "words": [
    {"word": "Hello", "start": 0.02, "end": 0.31, "confidence": 0.99},
    {"word": "from",  "start": 0.32, "end": 0.55, "confidence": 0.98},
    {"word": "slng",  "start": 0.56, "end": 1.10, "confidence": 0.96}
  ],
  "language": "en"
}

Streaming returns a series of events: ready, partial_transcript, final_transcript, error.

Audio formats

• File: WAV, MP3, FLAC, OGG, WebM, M4A, AAC
• Streaming: raw PCM linear16 at 16 kHz, mono, sent in 4096-byte chunks

Error handling

See also

• references/installation.md
• references/streaming.md
• references/file-transcription.md