feat: add MiniMax TTS provider support

[octo-patch]

Summary

This PR adds MiniMax TTS support to coqui-ai/TTS via a lightweight, standalone MiniMaxTTS class.

What's added

• TTS/minimax_tts.py — a new MiniMaxTTS class that:

  • Calls the MiniMax T2A v2 API for synthesis
  • Streams SSE events and decodes hex-encoded MP3 audio fragments
  • Returns a float32 numpy array at 32 kHz, fully compatible with coqui-ai/TTS audio utilities (save_wav, etc.)
  • Supports speech-2.8-hd (default) and speech-2.8-turbo models
  • Exposes six English voice IDs out of the box
  • Reads the API key from the MINIMAX_API_KEY environment variable

• tests/tts_tests/test_minimax_tts.py — comprehensive test suite:

  • Unit tests (mock-based, no API key required)
  • Integration tests (auto-skipped when MINIMAX_API_KEY is not set)

Usage

from TTS.minimax_tts import MiniMaxTTS

# Requires MINIMAX_API_KEY in environment
tts = MiniMaxTTS()                              # default: speech-2.8-hd
wav = tts.tts("Hello, world!")                  # returns numpy float32 array
tts.tts_to_file("Hello!", file_path="out.wav")  # saves as WAV

# Custom model / voice
tts_turbo = MiniMaxTTS(model="speech-2.8-turbo")
wav = tts_turbo.tts("Quick synthesis!", voice="English_Lucky_Robot")

API reference

• TTS: https://platform.minimax.io/docs/api-reference/speech-t2a-http

Dependencies

No new dependencies are introduced. The MP3→numpy conversion uses torchaudio (already in requirements.txt) with a fallback to soundfile.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[eginhard]

Development is now at https://github.com/idiap/coqui-ai-TTS. However, Coqui TTS is for open-source models, not for accessing commercial APIs.