improved text & audio processing

[ther3zz]

• better sentence splitting
• normalization for different symbols

I noticed that sentences similar to "you can expect sunny skies with a high of 73°F and a low of 54°F" would not be spoken properly when it came to the degrees F, figured I would work on enhancing that bit.

[ther3zz]

added the following for audio processing:

Cache Configuration:

• AUDIO_CACHE_MAX_SIZE_MB: Max in-memory cache size in MB. Evicts old entries when full. (Default: 256)
• AUDIO_CACHE_CLEAR_INTERVAL_S: Automatically clear cache periodically (in seconds). 0 disables. (Default: 3600)

Performance & Limits:

• AUDIO_SILENCE_PADDING_MS: Default silence duration in ms. (Default: 250)
• AUDIO_MAX_FILES_TO_CONCATENATE: Max number of files per job. (Default: 5000)
• AUDIO_MAX_TOTAL_SIZE_MB: Max combined file size in MB. (Default: 2048)
• AUDIO_USE_PARALLEL_PROCESSING: Set 'true' or '1' to enable parallel mode. (Default: false)
• AUDIO_MAX_PARALLEL_WORKERS: Max threads for parallel mode. (Default: CPU cores)
• AUDIO_LARGE_FILE_THRESHOLD_MB: Warn if a single file exceeds this size. (Default: 100)

Cache hit scenario: Near-instant (just memory copy)
Parallel cold start: ~0.5-2 seconds per 100 files
Sequential cold start: ~2-8 seconds per 100 files
Memory usage: Bounded by cache + parallel worker limits

TTS Use Case Optimization 🎤
Common phrases get cached (greetings, numbers, etc.)
Parallel processing handles large documents efficiently
Silence padding creates natural speech rhythm
Normalization ensures consistent volume
Multiple format support for different deployment targets