Skip to content

Changelog

0.1.0

Initial release.

  • STTClient with 4 transport protocols: post(), stream(), sse(), ws()
  • TTSClient with 4 transport protocols: post(), stream(), sse(), ws()
  • STTSession with built-in latency tracking, fragment accumulation, and auto-metrics
  • STTResult with chainable compute_metrics() and assert_quality()
  • TextMetrics — WER/CER via jiwer with detailed edit distance breakdown
  • AudioMetrics — PESQ MOS via pesq for perceptual audio quality
  • SampleRegistry — auto-discovery of ground-truth WAV + TXT pairs
  • Embedded samples — 3 English ground-truth recordings (hello_world, quick_brown_fox, counting)
  • audioeval fixture — session-scoped, CLI-configured facade
  • audioeval_thresholds fixture — per-test quality gates from CLI options
  • CLI options: --stt-url, --tts-url, --audioeval-wer, --audioeval-cer, --audioeval-mos
  • Full PEP 561 typing with py.typed marker