v0.2.0

v0.2.0 — Real-time mic → WAV, with VAD segments and cleaner NR ✨🎤

• What’s new

  • Millisecond-first API: Simple timing controls; safe integer-millisecond buffering ⏱️
  • Stream-to-file: Continuous 16‑bit WAV writing with periodic processing 💾
  • Optional VAD: Emit speech segments to a queue and/or save per-segment WAVs 🗣️➡️📂
  • Noise reduction first: NR applied before gain for consistent loudness 🔇➜🔊
  • Observability: Switched to logging for friendly, toggleable output 🧭

• Breaking changes

  • Renamed keep_before_speech_ms → pre_speech_padding_ms, keep_after_speech_ms → post_speech_padding_ms for clarity (intuitive naming) [[memory:6212612]] ✅
  • Removed VAD-specific sample_rate/buffer_size (VAD shares audio settings) 🧹
  • Signature: input_audio(output_audio_filepath, ...) is now positional for output; returns b"" (side-effects are the file and VAD outputs) ⚠️

• Key API

  • AudioConfig: sample_rate=16000, channels=1, buffer_size=512, batch_process_ms=320, gain_db=20.0
  • VADConfig: threshold=0.5, pre_speech_padding_ms=300, post_speech_padding_ms=500
  • NoiseReductionConfig: prop_decrease=0.8, etc.
  • Constraints enforced:
    • buffer_size * 1000 % sample_rate == 0
    • batch_process_ms is a multiple of buffer duration
    • AudioConfig.sample_rate == NoiseReductionConfig.sample_rate

• Tiny example

from input_audio import input_audio, VADConfig

input_audio(
    "./recordings/full.wav",
    enable_vad=True,
    vad_config=VADConfig(pre_speech_padding_ms=300, post_speech_padding_ms=500),
    vad_dirpath="./tmp_vad",  # optional per-segment WAVs
    max_recording_duration_ms=10_000,
)

• Quality improvements

  • Fade-in/out at segment boundaries to reduce clicks 🎛️
  • Accurate float32 pipeline; robust int16 conversion for WAV 📈
  • Final flush ensures no frames left behind on exit ✅
  • Safe gain with clipping protection 🔧

• Dependencies

  • Added: pydantic, durl-py, explicit numpy, torch, torchaudio, logging-bullet-train
  • Removed: soundfile
  • Bumped many versions; see requirements*.txt and pyproject.toml 📦

• Docs & meta

  • README rewritten: Quick Start, API reference, constraints, and examples 📝
  • Version bump: 0.1.0 → 0.2.0
  • Dev: added .envrc, updated .gitignore (tmp**) 🧰

• Reviewer tips

  • Focus on input_audio/__init__.py: new streaming loop, periodic processing, VAD emission, and config models
  • Try:

from input_audio import input_audio
input_audio("./recordings/quick.wav", max_recording_duration_ms=5000)

• Optional VAD run will write segments into ./tmp_vad and/or queue 🎯

• Why this change

  • Cleaner, predictable timing; clearer names; better defaults; safer audio pipeline; easier integration in streaming apps 🚀

What's Changed

• Dev/v0.2.0 by @allen2c in #1

New Contributors

• @allen2c made their first contribution in #1

Full Changelog: v0.1.0...v0.2.0