Improve TTFB of Deepgram TTS

[adithyaxx]

Issue

Deepgram's Python SDK produces slow TTS outputs with ~2.1s average TTFB.

Solution

Switch to direct HTTP POST requests, reducing average TTFB to ~0.8s (more than 2x faster).

[codecov]

Codecov Report

❌ Patch coverage is 0% with 23 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/pipecat/services/deepgram/tts.py	0.00%	23 Missing ⚠️

Files with missing lines	Coverage Δ
src/pipecat/services/deepgram/tts.py	0.00% <0.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[aconchillo]

Issue

Deepgram's Python SDK produces slow TTS outputs with ~2.1s average TTFB.

Solution

Switch to direct HTTP POST requests, reducing average TTFB to ~0.8s (more than 2x faster).

Wow! I don't think we'll have time to take a look at this before the upcoming release, but very interesting. I'm wondering if we are doing something wrong in our current code.

[adithyaxx]

@aconchillo I feel the current implementation is correct since it follows Deepgram's documentation. However, comparing it to direct API calls shows a difference in TTFB, suggesting the additional delays are coming from Deepgram's Python SDK itself. This finding came from comparing the implementation in World’s Fastest Talking AI: Deepgram + Groq with Pipecat's implementation.

[markbackman]

Awesome find! I implemented a version that I think is ready for production here:
#2943

If you want to take the changes from #2943 and make them here, that would be great. Alternatively, we could just close this PR, but I want you to be able to take credit for this one, if you'd like.

[markbackman]

@adithyaxx we've taken this change but added a new DeepgramHttpTTSService class. I'm closing this PR since the new PR now have a fair amount of changes. Again, thanks so much for bringing this up!