transcribe() - RunAnywhere Documentation

import com.runanywhere.sdk.public.RunAnywhere
import com.runanywhere.sdk.public.api.*

val transcription = RunAnywhere.stt.transcribe(AudioInput.file(path))
println(transcription.text)

Signature

suspend fun transcribe(audio: AudioInput, options: SttOptions? = null): Transcription

Building an AudioInput

AudioInput.pcm16(bytes: ByteArray, sampleRate: Int = 16000, channels: Int = 1)
AudioInput.float32(samples: FloatArray, sampleRate: Int = 16000, channels: Int = 1)
AudioInput.float32Bytes(bytes: ByteArray, sampleRate: Int = 16000, channels: Int = 1)
AudioInput.wav(bytes: ByteArray)
AudioInput.file(path: String)

From a recording:

val transcription = RunAnywhere.stt.transcribe(AudioInput.file(file.absolutePath))

From AudioRecord samples:

val buffer = ShortArray(bufferSize)
recorder.read(buffer, 0, buffer.size)

val bytes = ByteBuffer.allocate(buffer.size * 2)
    .order(ByteOrder.LITTLE_ENDIAN)
    .apply { asShortBuffer().put(buffer) }
    .array()

val transcription = RunAnywhere.stt.transcribe(AudioInput.pcm16(bytes))

float32Bytes is the raw-bytes form of float32, for when you already hold little-endian float samples rather than a FloatArray.

With options

val transcription = RunAnywhere.stt.transcribe(
    AudioInput.file(path),
    SttOptions(language = "en", punctuation = true, wordTimestamps = true)
)

See STT options.

Checking readiness

val state = RunAnywhere.stt.state()

Permission

Recording needs RECORD_AUDIO at runtime:

if (ContextCompat.checkSelfPermission(context, Manifest.permission.RECORD_AUDIO)
    != PackageManager.PERMISSION_GRANTED) {
    launcher.launch(Manifest.permission.RECORD_AUDIO)
}