Skip to content

Dictation: 44.1 kHz stereo WAV is 3-5x slower than 16 kHz mono; docs don't recommend downsampling #219

Open

Description

@YashasviThakur

Dictation API feedback (no public repo for it, so filing here since this SDK is the main client).

Same 8.25 s utterance sent to /v1/transcribe/live with default config, three runs each (server-reported request_time_ms):

Audio Run 1 Run 2 Run 3
16 kHz mono s16 WAV 2,092 ms 1,161 ms 1,010 ms
44.1 kHz stereo s16 WAV 9,796 ms 5,079 ms 3,734 ms

sync_time_ms (the transcription part) shows the same gap (about 0.9-2.0 s vs 3.2-8.3 s), so it is not the rewrite. Browsers and most microphones capture at 44.1/48 kHz, and "typical short clips respond in under a second" on the error-handling page is only true for 16 kHz mono. The audio-requirements page lists formats but never recommends a sample rate.

Asks: document "send 16 kHz mono for lowest latency", and/or make server-side resampling cheaper. The SDK could also downsample or warn.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions