Add the Telnyx vendor

[fmv1992]

Description

This PR introduces support for Telnyx Speech-to-Text (STT) and Text-to-Speech (TTS) services, enabling real-time conversational AI pipelines using Telnyx's voice APIs.

The implementation focuses on low latency and minimal dependencies by leveraging native PCM streaming.

Key Features

• TelnyxSTTService: Supports streaming raw audio (WAV/PCM) to Telnyx for real-time transcription.
• TelnyxTTSService: Supports streaming raw PCM audio from Telnyx (24kHz by default for NaturalHD voices), bypassing the need for external decoding libraries like ffmpeg or pydub.
• src/pipecat/services/telnyx/utils.py: Added utility for generating valid WAV headers required by the Telnyx streaming protocol.
• Example Agent: Added examples/foundational/telnyx_openai_agent.py demonstrating a complete voice-to-voice agent using Telnyx + OpenAI + Daily.co.

Implementation Details

• Protocol: Switched from mp3 to pcm in the WebSocket handshake. This eliminates decoding overhead and external dependencies, significantly reducing Time-to-First-Byte (TTFB).
• Sample Rates:
  • STT input defaults to 16kHz (optimal for VAD/ASR).
  • TTS output defaults to 24kHz (native rate for Telnyx NaturalHD voices) to ensure high-fidelity playback without pitch shifting.
• Context Aggregation: The example agent implements LLMUserResponseAggregator and LLMAssistantResponseAggregator to handle conversational turns robustly, including handling race conditions where VAD silence detection might outpace cloud transcription.

Dependencies

• No new OS-level dependencies (e.g., ffmpeg is not required).
• Standard Python websockets and aiohttp.

How to Test

1. Set the following environment variables:

   export TELNYX_API_KEY="your_key"
   export DAILY_ROOM_URL="your_daily_url"
   export DAILY_ROOM_TOKEN="your_daily_token"
   export OPENAI_API_KEY="your_openai_key"

2. Run the foundational example:

   python3 examples/foundational/telnyx_openai_agent.py

[sonamg-droid]

@aconchillo Hi Aleix, we’d appreciate a review from someone on the team so this PR can be approved and merged.

We’re contributing this from Telnyx to add official support for Telnyx Speech-to-Text and Text-to-Speech services, which we think can help broaden the range of real-time voice AI pipelines Pipecat supports.

Happy to make any changes based on feedback.