Fix Riva TTS

[openminddev]

This pull request introduces several improvements to audio processing and logging behavior, as well as adjustments to default parameters for both ASR and TTS modules. The most notable changes are enhancements to ASR sample rate handling, improved logging for ASR results, and streamlined TTS synthesis calls.

Audio Processing and ASR Handling

• Improved ASR sample rate initialization: The ASR processor now waits for the first audio chunk to dynamically set the sample rate and re-initialize the model if needed, ensuring compatibility with incoming audio streams. (src/om1_speech/riva/asr_processor.py)
• Refined default ASR thresholds and silence history: Default values for stop_history_eou, stop_threshold_eou, stop_history, and stop_threshold are set to achieve better balance between responsiveness and avoiding mid-sentence cuts. (src/om1_speech/riva/asr_processor.py)
• Changed default ASR sample rate argument from 16kHz to 48kHz for higher audio fidelity. (src/om1_speech/riva/args.py)

Logging and Diagnostics

• Enhanced ASR result logging: Interim and final ASR results are now logged separately for clearer diagnostics. (src/om1_speech/riva/asr_processor.py)
• Logging configuration now forcibly resets all handlers on initialization to ensure consistent log output. (src/om1_speech/main.py)

TTS Module Simplification

• Simplified TTS synthesis call: Removed unused parameters (audio_prompt_file, quality, custom_dictionary) for a cleaner interface and updated documentation accordingly. (src/om1_speech/riva/tts_processor.py) [1] [2]

WebSocket Server Responsiveness

• Reduced sleep interval for message processing in WebSocket server to improve responsiveness. (src/om1_utils/ws/server.py) [1] [2]