STT Options

RASTTOptions controls language, timestamps, diarization, and decoding behavior. It is a generated protobuf message: start from .defaults() and assign the fields you need.

var options = RASTTOptions.defaults()
options.language = .en
options.enableWordTimestamps = true

Defaults

.defaults() sets three fields; everything else stays at the proto zero value.

Field	Default	Notes
`language`	`.en`	`RASTTLanguage` enum
`enablePunctuation`	`true`
`enableWordTimestamps`	`true`

All fields

Field	Type	Description
`language`	`RASTTLanguage`	Target language, or `.auto` to detect
`languageCode`	`String`	BCP-47 override, guarded by `hasLanguageCode`
`detectLanguage`	`Bool`	Ask the model to detect the language
`enablePunctuation`	`Bool`	Insert punctuation and casing
`enableWordTimestamps`	`Bool`	Populate `output.words`
`enableDiarization`	`Bool`	Attribute segments to speakers
`maxSpeakers`	`Int32`	Upper bound for diarization
`vocabularyList`	`[String]`	Terms to bias decoding toward
`beamSize`	`Int32`	Beam search width, 0 leaves the backend default
`audioFormat`	`RAAudioFormat`	`.pcm`, `.wav`, `.mp3`, `.opus`, and others
`sampleRate`	`Int32`	Input sample rate in Hz
`maxAlternatives`	`Int32`	Number of alternative hypotheses to return
`chunkDurationMs`	`Int32`	Chunk size for streaming transcription
`endpointSilenceMs`	`Int32`	Silence before an endpoint is declared
`suppressBlank`	`Bool`	Drop blank decodes
`translateToEnglish`	`Bool`	Translate rather than transcribe

There is no enableVAD field on RASTTOptions. Voice activity detection is a separate component; see VAD.

Language

language is the RASTTLanguage enum, not a string. Cases: .unspecified, .auto, .en, .es, .fr, .de, .zh, .ja, .ko, .it, .pt, .ar, .ru, .hi. Two helpers bridge to BCP-47 text:

let language = RASTTLanguage.fromBcp47("es-MX")  // .es
let code = RASTTLanguage.ja.bcp47Code            // "ja"

Auto-detection

var options = RASTTOptions.defaults()
options.language = .auto

let output = try await RunAnywhere.transcribe(audio: audioData, options: options)
print("Detected: \(output.language.bcp47Code)")

Pin a language

var options = RASTTOptions.defaults()
options.language = .es

let output = try await RunAnywhere.transcribe(audio: audioData, options: options)

Word timestamps

Word-level timing drives subtitles, karaoke, and word highlighting. Times are in milliseconds.

var options = RASTTOptions.defaults()
options.enableWordTimestamps = true

let output = try await RunAnywhere.transcribe(audio: audioData, options: options)

for word in output.words {
    print("\(word.word): \(word.startMs)ms - \(word.endMs)ms")
}

RAWordTimestamp

Field	Type	Description
`word`	`String`	The word
`startMs`	`Int64`	Start offset in milliseconds
`endMs`	`Int64`	End offset in milliseconds
`confidence`	`Float`	Word-level confidence
`speakerID`	`String`	Speaker label, guarded by `hasSpeakerID`

Subtitle grouping

struct Subtitle {
    let text: String
    let startMs: Int64
    let endMs: Int64
}

func generateSubtitles(from audioData: Data) async throws -> [Subtitle] {
    var options = RASTTOptions.defaults()
    options.enableWordTimestamps = true

    let output = try await RunAnywhere.transcribe(audio: audioData, options: options)

    guard !output.words.isEmpty else {
        return [Subtitle(text: output.text, startMs: 0, endMs: output.durationMs)]
    }

    return stride(from: 0, to: output.words.count, by: 7).map { start in
        let chunk = output.words[start..<min(start + 7, output.words.count)]
        return Subtitle(
            text: chunk.map(\.word).joined(separator: " "),
            startMs: chunk.first!.startMs,
            endMs: chunk.last!.endMs
        )
    }
}

Diarization

Attribute each segment to a speaker. The labels land in output.speakerIds and on each RAWordTimestamp.

var options = RASTTOptions.defaults()
options.enableDiarization = true
options.maxSpeakers = 4

Sample rate

Whisper models expect 16 kHz mono. Set sampleRate when the source differs.

var options = RASTTOptions.defaults()
options.sampleRate = 44100

Record at 16 kHz mono where you can. Resampling costs accuracy.

Transcription metadata

RATranscriptionMetadata reports what ran and how long it took.

let output = try await RunAnywhere.transcribe(audio: audioData, options: options)

if output.hasMetadata {
    let metadata = output.metadata
    print("Model: \(metadata.modelID)")
    print("Processing time: \(metadata.processingTimeMs)ms")
    print("Audio length: \(metadata.audioLengthMs)ms")
    print("Real-time factor: \(metadata.realTimeFactor)x")
}

Alternative transcriptions

Set maxAlternatives to receive competing hypotheses in output.alternatives, each an RATranscriptionAlternative with text, confidence, and words.

var options = RASTTOptions.defaults()
options.maxAlternatives = 3

let output = try await RunAnywhere.transcribe(audio: audioData, options: options)

print("Best: \(output.text)")
for (index, alternative) in output.alternatives.enumerated() {
    print("Alt \(index + 1): \(alternative.text) (\(alternative.confidence))")
}

Complete example

struct TranscriptionResult {
    let text: String
    let confidence: Float
    let language: RASTTLanguage
    let words: [RAWordTimestamp]
    let processingTimeMs: Int64
}

func transcribe(
    audioURL: URL,
    language: RASTTLanguage = .en,
    includeTimestamps: Bool = false
) async throws -> TranscriptionResult {
    let audioData = try Data(contentsOf: audioURL)

    var options = RASTTOptions.defaults()
    options.language = language
    options.sampleRate = 16000
    options.enableWordTimestamps = includeTimestamps

    let output = try await RunAnywhere.transcribe(audio: audioData, options: options)

    return TranscriptionResult(
        text: output.text,
        confidence: output.confidence,
        language: output.language,
        words: output.words,
        processingTimeMs: output.hasMetadata ? output.metadata.processingTimeMs : 0
    )
}

transcribe()

Basic transcription

Streaming STT

Real-time transcription

Getting Started

Swift SDK

Kotlin SDK

React Native SDK

Flutter SDK

Web SDK

Vibe Coding

Defaults

All fields

Language

Auto-detection

Pin a language

Word timestamps

RAWordTimestamp

Subtitle grouping

Diarization

Sample rate

Transcription metadata

Alternative transcriptions

Complete example

transcribe()

Streaming STT

​Defaults

​All fields

​Language

​Auto-detection

​Pin a language

​Word timestamps

​RAWordTimestamp

​Subtitle grouping

​Diarization

​Sample rate

​Transcription metadata

​Alternative transcriptions

​Complete example

transcribe()

Streaming STT

Defaults

All fields

Language

Auto-detection

Pin a language

Word timestamps

RAWordTimestamp

Subtitle grouping

Diarization

Sample rate

Transcription metadata

Alternative transcriptions

Complete example