Gemini Transcribe Speech to Text in Node.js
GeminiSTT streams the voice of the user to Gemini Transcribe over the Live
API, and emits the transcript of each utterance a fraction of a second after it ends.
Installation
npm install @micdrop/geminiThe same package also provides an agent, a text to speech, Gemini Live and a classifier. The providers page shows what every provider covers.
Usage with MicdropServer
import { GeminiSTT } from '@micdrop/gemini'import { MicdropServer } from '@micdrop/server'
const stt = new GeminiSTT({ apiKey: process.env.GEMINI_API_KEY || '', language: 'en-US', // Optional, detected when left out vocabulary: ['Micdrop'], // Optional, words to recognize})
new MicdropServer(socket, { stt, // ... other options})Usage without MicdropServer
import { GeminiSTT } from '@micdrop/gemini'import { createReadStream } from 'fs'
const stt = new GeminiSTT({ apiKey: process.env.GEMINI_API_KEY || '' })
stt.on('Transcript', (transcript) => console.log('Transcript:', transcript))
// Audio is raw PCM, 16 bits, 16 kHz, monostt.transcribe(createReadStream('speech.pcm'))Events
| Event | Payload | Description |
|---|---|---|
Transcript | string | The transcript of an utterance, empty when nothing was understood. |
Failed | Buffer[] | The connection could not be restored, with the audio still untranscribed. |
See the STT interface for the full contract.
Each stream given to transcribe() is one utterance with a transcript of its
own, and the transcripts come in the order of the streams. A stream can start
while the previous one still waits for its transcript: Gemini names no
utterance in what it sends, so it hears them one after the other, and the audio
of the next one waits for the transcript of the previous.
Options
| Option | Type | Default | Description |
|---|---|---|---|
apiKey | string | Required | Your Gemini API key |
model | string | 'gemini-3.5-transcribe-live' | Transcription model |
language | string | undefined | Language of the user, as a BCP-47 code like fr-FR |
vocabulary | string[] | undefined | Words to recognize, up to 1,000 |
mode | 'SMART' | 'VERBATIM' | Model default | Cleans up hesitations, or keeps every word as said |
transcriptionTimeout | number | 4000 | How long to wait for a transcript, in milliseconds |
connectionTimeout | number | 5000 | Time to open the connection, in milliseconds |
retryDelay | number | 1000 | Delay before reconnecting, in milliseconds |
maxRetry | number | 3 | Reconnection attempts before giving up |