Streaming STT - RunAnywhere Documentation

Two streaming shapes exist. Pick by where your audio comes from.

Push: openStream

You hold a handle and push frames into it as the microphone produces them.

let stream = try await RunAnywhere.stt.openStream(
    format: AudioFormatSpec(encoding: .pcmF32Le, sampleRate: 16000, channels: 1)
)

// as audio arrives
stream.pushFrame(AudioFrame(samples: data, sampleCount: count))

// when the utterance ends
stream.flush()
stream.finish()
await stream.close()

SttStream exposes pushFrame(_:), flush(), finish(), and close().

Pull: transcribeStream

You already have an AsyncStream<AudioInput> and want events back.

let events = try await RunAnywhere.stt.transcribeStream(audioStream)

for try await event in events {
    switch event {
    case .partial(let text):
        preview = text
    case .final(let transcription):
        transcript += transcription.text
    default:
        break
    }
}

Signatures

func openStream(format: AudioFormatSpec, options: SttOptions? = nil) async throws -> SttStream
func transcribeStream(_ audio: AsyncStream<AudioInput>, options: SttOptions? = nil) async throws
    -> AsyncThrowingStream<TranscriptionEvent, Error>

AudioFormatSpec

Field	Type	Default
`encoding`	`AudioEncoding`	required
`sampleRate`	`Int`	required
`channels`	`Int`	`1`
`container`	`AudioFormat?`	none

Choosing between them

Push suits a live microphone: you already have a callback delivering buffers, so you push each one as it lands. Pull suits a source you can express as an AsyncStream, such as a file being read in chunks or audio arriving over the network.