STT Options

STTOptions is generated. Import it from @runanywhere/proto-ts/stt_options. Both transcribe and transcribeStream accept a Partial<STTOptions>.

Fields

Field	Type	SDK default	Notes
`language`	`STTLanguage`	`STT_LANGUAGE_EN`
`languageCode`	`string`	unset	Free-form BCP-47; preferred over the enum when set
`detectLanguage`	`boolean`	derived	True when `language` is unset or `STT_LANGUAGE_AUTO`
`enablePunctuation`	`boolean`	`true`
`enableWordTimestamps`	`boolean`	`true`
`enableDiarization`	`boolean`	`false`
`maxSpeakers`	`number`	`0`	0 is auto
`vocabularyList`	`string[]`	`[]`	Custom vocabulary bias
`beamSize`	`number`	`0`	0 is the backend default
`audioFormat`	`AudioFormat`	`AUDIO_FORMAT_PCM`	From `@runanywhere/proto-ts/model_types`
`sampleRate`	`number`	`16000`
`maxAlternatives`	`number`	`0`	0 is the backend default
`chunkDurationMs`	`number`	`0`	Streaming control; 0 is the backend default
`endpointSilenceMs`	`number`	`0`	Streaming control; 0 is the backend default
`suppressBlank`	`boolean`	`false`
`translateToEnglish`	`boolean`	`false`

The SDK fills the defaults above before encoding the request, so an omitted field is not the same as a proto zero.

Language

STTLanguage is an enum, not a string. Its members carry the two-letter wire tag.

import { STTLanguage } from '@runanywhere/proto-ts/stt_options'

const result = await RunAnywhere.transcribe(pcmBytes, {
  language: STTLanguage.STT_LANGUAGE_ES,
})

Member	Tag	Member	Tag
`STT_LANGUAGE_AUTO`	`auto`	`STT_LANGUAGE_KO`	`ko`
`STT_LANGUAGE_EN`	`en`	`STT_LANGUAGE_IT`	`it`
`STT_LANGUAGE_ES`	`es`	`STT_LANGUAGE_PT`	`pt`
`STT_LANGUAGE_FR`	`fr`	`STT_LANGUAGE_AR`	`ar`
`STT_LANGUAGE_DE`	`de`	`STT_LANGUAGE_RU`	`ru`
`STT_LANGUAGE_ZH`	`zh`	`STT_LANGUAGE_HI`	`hi`
`STT_LANGUAGE_JA`	`ja`	`STT_LANGUAGE_UNSPECIFIED`	none

For a regional variant set languageCode: 'en-GB'; consumers prefer it over the enum. Language-specific models such as whisper-tiny.en only handle their one language, and specifying it is both faster and more accurate than auto-detection.

Word timestamps

enableWordTimestamps is on by default. Timings arrive in STTOutput.words as WordTimestamp entries, not in a segments array.

const result = await RunAnywhere.transcribe(pcmBytes, {
  enableWordTimestamps: true,
})

for (const word of result.words) {
  console.log(word)
}

Diarization

const result = await RunAnywhere.transcribe(pcmBytes, {
  enableDiarization: true,
  maxSpeakers: 2,
})

console.log(result.speakerIds)

Diarization is expensive and backend-dependent. speakerIds stays empty when the loaded model does not support it.

Sample rate

16 kHz mono is what the STT frontends expect, and it is the SDK default for both the request and AudioCaptureManager. Passing a higher rate makes the backend downsample, which costs time and buys nothing.

Alternatives

const result = await RunAnywhere.transcribe(pcmBytes, { maxAlternatives: 3 })
for (const alt of result.alternatives) {
  console.log(alt)
}

Performance

Option	Speed	Accuracy
`language` pinned	Faster	Better
`enableWordTimestamps`	Slower	Unchanged
`enableDiarization`	Much slower	Unchanged
`enablePunctuation`	Roughly equal	Unchanged

Transcribe

Buffer transcription

STT Streaming

Live partial results

VAD

Voice activity detection

Getting Started

Swift SDK

Kotlin SDK

React Native SDK

Flutter SDK

Web SDK

Vibe Coding

Fields

Language

Word timestamps

Diarization

Sample rate

Alternatives

Performance

Transcribe

STT Streaming

VAD

​Fields

​Language

​Word timestamps

​Diarization

​Sample rate

​Alternatives

​Performance

​Related

Transcribe

STT Streaming

VAD

Fields

Language

Word timestamps

Diarization

Sample rate

Alternatives

Performance

Related