PlayHT shut down: move your PlayHT API calls to Cartesia

Move this codebase from PlayHT to Cartesia for text to speech. Keep PlayHT working behind a flag until I switch over.

Shut down on December 31, 2025. Meta confirmed it had acquired PlayAI, PlayHT's parent, on July 12, 2025, and the API went offline around July 26, 2025. Inworld and Murf both give December 31, 2025 as the date the platform closed for good. We found no shutdown notice from PlayHT itself. On November 5, 2026 (UTC), `play.ht` and `api.play.ht` no longer resolved.

July 12, 2025: Meta confirms it acquired PlayAI. Meta bought the company behind PlayHT and did not disclose terms. The team moved to Meta. (https://www.biometricupdate.com/202507/meta-acquires-voice-generator-startup-playai)

July 26, 2025: The PlayHT API goes offline. API calls stopped working about two weeks after the acquisition, months before the platform's announced end date. (https://inworld.ai/resources/migrate-from-playht)

## 1. Find the calls
Search for `api.play.ht`, `PLAYHT_API_KEY` or `PLAYHT_USER_ID`. The request to replace is `POST https://api.play.ht/api/v2/tts/stream`. List every call site before you edit anything.

## 2. Target
- Request: `POST https://api.cartesia.ai/tts/bytes`
- Model: `model_id: "sonic-3.6"`
- Headers: `X-API-Key` read from the `CARTESIA_API_KEY` environment variable, and `Cartesia-Version: 2026-08-14`
- Auth swap: PlayHT sends the key in `AUTHORIZATION: Bearer` secret key plus an `X-USER-ID` header. Replace it with `X-API-Key` from `CARTESIA_API_KEY` on the Cartesia path only, and keep PlayHT's key for the PlayHT path. Both values are dead now. Remove them from your secrets store.
- Add `CARTESIA_API_KEY` wherever the project documents its environment (`.env.example`, README, deploy config). Never hardcode the key.
- Request size: PlayHT caps a request at 2,000 characters per request; 20,000 on Play3.0-mini. Cartesia publishes no per-request maximum, so keep any chunking the code already does and split at sentence boundaries; for long text, stream over the Cartesia TTS WebSocket.
- Voice IDs don't carry over. Keep each voice ID in one named constant or config value, put a placeholder where each PlayHT voice was, and list them for me so I can pick replacements at https://www.cartesia.ai/voices.
- Suggested starting points to put in those placeholders, for me to confirm by ear: Podcast or audiobook narrator → `a33f7a4c-100f-41cf-a1fd-5822e8fc253f` (Lauren - Lively Narrator); Support line or IVR prompt → `f786b574-daa5-4673-aa0c-cbe3e8534c02` (Katie - Friendly Fixer); Measured male voice for explainers → `98a34ef2-2140-4c28-9c71-663dc4dd7022` (Clyde - Calm Narrator).

## 3. Field mapping, PlayHT → Cartesia
- `POST https://api.play.ht/api/v2/tts/stream` → `POST /tts/bytes`
- `AUTHORIZATION: Bearer <secret key>` and `X-USER-ID` → `X-API-Key` and `Cartesia-Version: 2026-08-14`. Delete the user ID from your config. Cartesia has no equivalent.
- `voice`: an `s3://` manifest URL → `voice`: `{"id": "<voice id>"}`. PlayHT voices and clones went offline with the service. Pick a library voice, or clone again from your original recording.
- `voice_engine`: `PlayDialog`, `Play3.0-mini`, `PlayHT2.0` and others → `model_id`: `sonic-3.6`
- `text` (2,000 characters; 20,000 on `Play3.0-mini`) → `transcript`
- `language`: a lowercase name such as `english` → `language`: a code such as `en` or `en-GB`
- `output_format`: `mp3`, `wav` or `mulaw` → `output_format` object: `container` `mp3`, `wav` or `raw`, with `pcm_mulaw` for mu-law
- `sample_rate` (8000 to 48000) → `output_format.sample_rate` (8000 to 48000)
- `speed` (0.1 to 5) → `generation_config.speed` (0.6 to 1.5). Anything slower than 0.6 or faster than 1.5 has to be clamped.
- `emotion` → `generation_config.emotion`. The two vendors use different emotion names. Map each one by ear.

## 4. PlayHT fields with no Cartesia equivalent
- `output_format`: `ogg` or `flac`. What to do: Generate `mp3` or PCM, then convert with ffmpeg or your encoder of choice.
- `quality`, `temperature`, `seed`, `voice_guidance`, `style_guidance`, `text_guidance`. What to do: Drop them.
- `voice_2`, `turn_prefix`, `turn_prefix_2` (two-speaker PlayDialog). What to do: Synthesize each speaker's turns with that speaker's voice and join the clips in order.

## 5. Reference request
Match the language and HTTP client the codebase already uses. This is the same request in Python:
```python
import os
import requests

text = "Episode twelve. Today we talk to a beekeeper about the first warm week of spring."
VOICE_ID = "6ccbfb76-1fc6-48f7-b71d-91ac6298247b"  # from /voices

response = requests.post(
    "https://api.cartesia.ai/tts/bytes",
    headers={
        "X-API-Key": os.environ["CARTESIA_API_KEY"],
        "Cartesia-Version": "2026-08-14",
    },
    json={
        "model_id": "sonic-3.6",
        "transcript": text,
        "voice": {"id": VOICE_ID},
        "language": "en",
        "output_format": {
            "container": "mp3",
            "sample_rate": 44100,
            "bit_rate": 128000,
        },
    },
)
response.raise_for_status()
open("out.mp3", "wb").write(response.content)
```

## 6. Roll out
1. Choose PlayHT or Cartesia per request from one config flag, defaulting to PlayHT.
2. Run the same scripts through both and save the clips side by side for me to listen to. Check that the container, encoding and sample rate match what downstream code expects.
3. Leave the PlayHT client, call and key in place so rollback is flipping the flag. Don't delete them in this change.
4. Run the existing tests, and add one that checks the Cartesia request body.

## Docs
- PlayHT stream API reference (docs.play.ht, still online): https://docs.play.ht/reference/api-generate-tts-audio-stream
- Inworld, migrating from PlayHT: https://inworld.ai/resources/migrate-from-playht
- Murf, PlayAI transition guide: https://murf.ai/blog/play-ai-transition-guide
- Cartesia TTS bytes API reference: https://docs.cartesia.ai/api-reference/tts/bytes
- Cartesia voice cloning guide: https://docs.cartesia.ai/build-with-cartesia/capability-guides/clone-voices

When you're done, tell me which files changed, which fields you couldn't map, and how to flip the flag.