Like #10 but for audio models.
There are now some audio models in Bedrock like Mistral AI Voxtral that can be used to implement audio routes (Like transcription for Voxtral).
The current Polly/Transcribe behavior will be a specific implementation of this.
