Fix macOS renderer microphone capture for Azure Speech

[RahulSinghParmar]

Summary

Fixes microphone capture on macOS when using the Azure Speech provider.

Previously, the application attempted to use native microphone capture on macOS. This caused the renderer microphone pipeline to not provide audio correctly.

Changes

• Use renderer-based microphone capture on macOS and Windows.
• Keep native microphone capture for Linux.
• Create the Azure PushStream before starting renderer audio capture.
• Forward renderer PCM audio to the Azure Speech PushStream.
• Start the renderer recording notification only after the Azure audio pipeline is initialized.
• Add logging around renderer microphone permission and PCM transmission.
• Preserve the existing Whisper renderer capture path.

Testing

Tested locally on macOS.

Observed:

• Azure Speech service initialized successfully.
• Azure recognition session started successfully.
• Renderer microphone permission granted.
• Renderer PCM audio received by the main process.
• Azure speech transcription successfully generated text.
• Gemini LLM responses were successfully generated from transcribed speech.

Example logs:

[AUDIO-IPC] Received renderer PCM
[SPEECH] Recognition session started
[SPEECH] Continuous Azure speech recognition started successfully
[MAIN] Processing transcription with intelligent LLM response

Platform

• macOS
• Azure Speech provider
• Renderer Web Audio API microphone capture

[vercel]

@RahulSinghParmar is attempting to deploy a commit to the csrsoftwares' projects Team on Vercel.

A member of the Team first needs to authorize it.