Speech gate

[pragmatrix]

Azure speech detection seems to be very sensitive. It detects very faint speech which makes it to also detect echos from speakers, for example when speech detection is active and someone else speaks on the other line.

This PR adds a speech gate that is loosely configured to dampen low volume audio, noise and also whispering sounds. The algorithm was mostly done by Claude Sonnet 3.7 and the parameterization was derived from iterating on two samples (normal speech and echo speech).

[Copilot]

Pull Request Overview

This PR implements a "speech gate" to address Azure speech detection sensitivity issues by filtering out low-volume audio, noise, and echo detection. The gate uses configurable parameters to dampen unwanted audio while preserving normal speech.

Key changes:

• Adds a new speech gate module with RMS-based filtering algorithm
• Integrates the speech gate into Azure transcription service
• Includes a standalone test utility for audio processing validation

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
core/src/speech_gate.rs	Implements the main speech gate algorithm with multiple variants (hard/soft/RMS-based)
core/src/lib.rs	Exports the speech gate module
src/lib.rs	Exports the speech gate processor function
services/azure/src/transcribe.rs	Integrates speech gate into Azure transcription pipeline
filter-test/	Adds a standalone CLI tool for testing speech gate on audio files
Cargo.toml	Adds fundsp dependency and filter-test workspace member

Comments suppressed due to low confidence (1)

filter-test/Cargo.toml:4

• The Rust edition "2024" does not exist. The latest stable edition is "2021". Change this to "2021".

edition = "2024"