flutter_nnnoiseless 1.4.0

Recurrent neural network for audio noise reduction .

1.4.0

• Web support: NoiselessSession (real-time denoising and VAD) now runs in the browser via bundled WebAssembly; no extra build steps, but apps must be served with cross-origin isolation headers (see the README Web section). denoiseFile and pcmToWav throw UnsupportedError on web. Adds ~760KB of wasm assets to app bundles.
• Example app runs on web (self-test and live microphone VAD demos)
• Dropped the unused rand dependency; upgraded wasm-bindgen to 0.2.126

1.3.0

• Denoise any audio file: denoiseFile now decodes FLAC, MP3, OGG/Vorbis, and M4A/AAC in addition to WAV (any bit depth), via symphonia; output remains 16-bit WAV at the input's rate and channel count
• denoiseFile feature parity with sessions: new wet (suppression strength) and model (custom RNNoise weights) parameters
• Corrupt packets in compressed files are skipped instead of failing the whole file; unsupported formats fail with a clear error
• Example app rewritten to showcase the API: file denoising with a live progress bar and cancel button, and a real-time microphone demo with a voice-activity meter
• Decode hardening from an adversarial review pass:
  • the decoder's actual output spec is trusted over container-declared values, so mis-remuxed or crafted files can no longer produce pitch-shifted or channel-garbled output while reporting success
  • crafted headers (0Hz or absurd sample rates, huge channel counts, unbounded stream lengths) now fail with clear errors instead of panicking or exhausting memory
  • a mid-file spec change or a long run of corrupt packets is an error instead of silent truncation; the first decodable audio track is selected even when the container's default track is not audio

1.2.0

• Multi-channel sessions: NoiselessSession.create(channels: n) denoises interleaved PCM with up to 8 channels, each with independent RNNoise state; channel alignment survives arbitrary chunk sizes
• Custom RNNoise models: pass model: bytes in the nnnoiseless training format to NoiselessSession.create to use your own trained weights
• File progress and cancellation: denoiseFile accepts onProgress (fraction 0.0 to 1.0) and cancelToken (throws DenoiseCancelledException on cancel)
• Precompiled binaries now include Linux ARM64 (aarch64-unknown-linux-gnu)
• CI now builds the example app for Linux desktop
• Hardening from an adversarial review pass:
  • sessions survive chunks split at odd byte offsets (a one-byte carry keeps 16-bit sample pairs and channel alignment intact)
  • NoiselessCancelToken can safely be constructed before any other plugin call (the native token is created lazily); tokens are one-shot and documented as such
  • cancellation is detected from the token state, never by matching the error message
  • denoiseFile reports progress and honors cancellation during the resample and write phases too, cleans up a partially-written output file on cancel, no longer panics on WAVs whose sample count is not a multiple of the channel count, and surfaces Rust panics as errors instead of silent success
  • a throwing onProgress callback now cancels the native work
  • whole-file denoising runs on a dedicated thread so it cannot starve real-time sessions; session creation moved off the UI thread
• Note for implementers extending Noiseless: denoiseFile gained two optional named parameters, so overrides must be updated to match

1.1.0

• New NoiselessSession API for real-time denoising: instance-based (run multiple concurrent streams), with process, flush, reset and dispose, plus a StreamTransformer (session.transformer) for piping a microphone stream straight through the denoiser
• Voice activity detection: process returns the RNNoise speech probability for every 10ms frame (DenoiseResult.voiceProbabilities, voiceProbability, isVoice())
• Output at your sample rate: sessions resample internally (streaming, artifact-free via rubato) and return audio at the input rate instead of always 48kHz
• Dry/wet mix: wet parameter (0.0-1.0) controls suppression strength
• denoiseFile robustness: supports 16/24/32-bit int and 32-bit float WAV input (previously crashed on anything but 16-bit int) and preserves the input sample rate in the output file
• RNNoise's warm-up frame is silenced instead of leaking into the output
• Deprecated Noiseless.denoiseChunk in favor of NoiselessSession
• Upgraded flutter_rust_bridge to 2.12.0
• Example app updated to the session API with a live voice-probability log

1.0.2

• Fixed denoiseChunk crashing with a stack overflow: the Dart wrapper was recursively calling itself instead of the underlying Rust binding, so real-time denoising never worked in 1.0.1
• Removed leftover placeholder comments from the public wrapper
• Fixed iOS/macOS podspec metadata (placeholder summary/description, broken homepage URL, stale 0.0.1 version) and aligned the macOS deployment target with the documented 10.15 minimum
• Corrected the meaningless flutter: ">=1.10.0" SDK constraint to >=3.0.0

1.0.1

• Changed method denoiseInRealtime to denoiseChunk for readability
• Added description to the denoise methods
• Removed rust_builder
• Updated README.md

1.0.0

• Initial release