Add FalSTTService

[markbackman]

Please describe the changes in your PR. If it is addressing an issue, please reference that as well.

Draft only. This is not in shape to be merged. Fal's service returns false positives with no audio, like this:

2025-02-11 12:48:06.817 | DEBUG    | pipecat.services.fal:run_stt:268 - Processing audio segment: 14124 bytes
2025-02-11 12:48:07.302 | DEBUG    | pipecat.services.fal:run_stt:289 - Full Fal response: {'text': 'you', 'chunks': [{'timestamp': [0.0, 0.441], 'text': 'you'}]}
2025-02-11 12:48:07.302 | DEBUG    | pipecat.services.fal:run_stt:294 - Transcription: [you]
2025-02-11 12:48:08.417 | DEBUG    | pipecat.services.fal:run_stt:268 - Processing audio segment: 9644 bytes
2025-02-11 12:48:08.729 | DEBUG    | pipecat.services.fal:run_stt:289 - Full Fal response: {'text': 'Thank you.', 'chunks': [{'timestamp': [0.0, 0.301], 'text': 'Thank you.'}]}
2025-02-11 12:48:18.012 | DEBUG    | pipecat.services.fal:run_stt:294 - Transcription: [Thank you.]
2025-02-11 12:48:19.937 | DEBUG    | pipecat.services.fal:run_stt:268 - Processing audio segment: 12844 bytes
2025-02-11 12:48:20.167 | DEBUG    | pipecat.services.fal:run_stt:289 - Full Fal response: {'text': 'Bye.', 'chunks': [{'timestamp': [0.0, 0.401], 'text': 'Bye.'}]}
2025-02-11 12:48:20.167 | DEBUG    | pipecat.services.fal:run_stt:294 - Transcription: [Bye.]

Once this blocker is resolved, we can proceed.

[phamtrung0633]

Hi Mark, this is also an issue (really a big issue) with Groq Whisper..

[markbackman]

Hi Mark, this is also an issue (really a big issue) with Groq Whisper..

Yup. You're right. That's really unfortunate, but Groq Whisper is so fast and affordable. I guess you can't have it all. Looks like it's just a Whisper problem.

This is telling: openai/whisper#679.