CHINESE STREAMING ASR

Live Lab

DETERMINISTIC REPLAY
Runtime
Pages / deterministic replay
Audio handling
Ephemeral memory
Evidence scope
Interactive sample
IDLE 00:00.0
No active override
First partial—onset → text
Endpoint → final—verifier path
RTF—compute / audio
Stability—fixed v1 definition
SIGNAL

Waveform & VAD

SILENCE
16 kHz · mono · PCM16LERMS — · peak —
No audio in the current session.
REVISION LOG

Self-healing transcript

0 revisions
PARTIAL—
    SESSION

    Timeline

    Start a demo session to populate the timeline.
    CONTROLLED CHALLENGES

    Robustness Arena

    Preview a declared channel recipe against the deterministic fixture. This page does not run transformed audio through ASR.

    CHALLENGE

    Signal recipe

    FIXTURE READY

    Fixture preview only. Reference text is absent, so CER remains N/A.

    RESULT

    Paired observation

    NOT RUN
    Condition—
    Fixture streaming text—
    Fixture verifier text—
    CER (reference)N/A
    Artifact—
    MATRIX

    Frozen benchmark boundary

    NOT YET EVALUATED
    No public robustness metric is displayed until a frozen manifest and independent recomputation are available.
    AUDITABLE PLAYBACK

    Session Timeline

    Every segment keeps its source, revision, and time span. Export contains provenance fields.

    EVENTS

    No session selected

      REPRODUCIBILITY LEDGER

      Evidence

      Claims stay attached to the model, protocol, device, and artifact that produced them.

      RUNTIME

      Interactive sample

      Backend
      deterministic-protocol-fixture
      Protocol
      echoforge.ws/v1
      Audio
      16 kHz · mono · PCM16LE
      Retention
      ephemeral memory
      EVALUATION

      Frozen benchmark

      Dataset
      AISHELL-1 / OpenSLR 33
      Normalizer
      echoforge.zh-normalizer/v1
      CER
      not yet evaluated
      RTF / p95
      not yet evaluated
      CLAIM BOUNDARY

      What this page proves

      • Protocol and state transitions are observable.
      • Revision provenance is preserved.
      • Deterministic replay is not a speech benchmark.
      • Decoder score is not calibrated probability.