Treatment transcribeContinuous

ml/remote/stt::transcribeContinuous

Configuration

⬡ stt: ml/remote/stt::RemoteStt

Inputs

⇥ segments: Stream<Vec<byte>>

Outputs

↦ error: Stream<string>
↦ failed: Stream<void>
↦ transcript: Stream<string>

Continuously transcribe a stream of audio segments to text.

For each Vec<byte> received on segments, submits it as a complete, independently decodable audio clip to the configured provider and emits the resulting text on transcript. The treatment runs until segments is closed.

Each item on segments must be a self-contained audio blob — the caller is responsible for producing meaningful segment boundaries (e.g. one WAV blob per utterance from a voice-activity detector, one chunk per WebSocket message from a streaming API, etc.). Splitting at the raw byte level within an encoded format would corrupt frames and produce garbage transcripts.

If a single segment fails, failed and error are emitted for that segment and processing continues with the next one.

graph LR
     T("transcribeContinuous()")
     S["🟩 🟩 🟩 …"] -->|segments|   T
     T -->|transcript| R["🟩 🟩 🟩 …"]
     T -->|failed|     F["🟩 🟩 🟩 …"]
     T -->|error|      E["🟩 🟩 🟩 …"]

     style S fill:#ffff,stroke:#ffff
     style R fill:#ffff,stroke:#ffff
     style F fill:#ffff,stroke:#ffff
     style E fill:#ffff,stroke:#ffff

use ml/remote/stt::RemoteStt
use ml/remote/stt::transcribeContinuous

treatment example()
  model stt: RemoteStt(backend = "openai", api_key = "sk-...", model = "whisper-1")
  input  segments:   Stream<Vec<byte>>
  output transcript: Stream<string>
{
    transcribeContinuous[stt=stt]()
    Self.segments -> transcribeContinuous.segments,transcript -> Self.transcript
}