Synthesize speech with Deepgram Aura text-to-speech over the REST endpoint

Documented steps

1. Create a Deepgram API key in the console.
2. POST to https://api.deepgram.com/v1/speak with the voice selected via query string, e.g. ?model=aura-2-thalia-en for Aura-2 or ?model=aura-asteria-en for Aura-1.
3. Set header Authorization: Token <DEEPGRAM_API_KEY>. (Bearer is used only with JWT tokens.)
4. Set header Content-Type: application/json.
5. Send body {"text": "..."} — text is the only required field.
6. Optionally add query params encoding (default mp3), container (default wav), sample_rate (default 24000), bit_rate (default 48000) and speed (0.7-1.5, not yet supported in all languages).
7. For asynchronous delivery set callback=<https url> and optionally callback_method (default POST) instead of holding the connection open.
8. Optionally set tag for usage reporting labels and mip_opt_out=true to opt out of the model improvement program.
9. Read the response as a binary audio stream — bytes begin arriving as soon as the first chunk is ready, so playback can start before synthesis completes.
10. Use response headers dg-model-name, dg-char-count and dg-request-id for logging, cost attribution and support tickets.
11. Write the body to a file matching the requested container/encoding (.mp3, .wav, etc.).
12. Handle errors: 400 bad request, 413 when text exceeds the character limit, 429 when concurrency limits are exceeded.
13. Verify against the official docs: https://developers.deepgram.com/reference/text-to-speech-api/speak , https://developers.deepgram.com/docs/text-to-speech and https://developers.deepgram.com/docs/tts-models