I've been running a series of voice cloning and TTS benchmarks using Resemble AI's API and noticed several 4xx errors on older integration scripts. Upon investigation, their latest API version (`v2` as of this writing) has officially deprecated a number of endpoints from `v1`.
The primary changes I've documented are:
* **Deprecation of `/voices` (list all) endpoint:** The call to `GET /api/v1/voices/` no longer returns a full list. Access is now scoped to a project. You must use `GET /api/v1/projects/{project_id}/voices/`.
* **Changes to voice creation workflow:** The direct `POST /api/v1/voices/` endpoint is deprecated. Voice cloning is now explicitly project-bound via `POST /api/v1/projects/{project_id}/voices/`.
* **Synthesis endpoint hardening:** The `POST /api/v1/synthesize/` endpoint now enforces stricter parameter validation, particularly around `project_id` and `voice` ownership.
This has clear implications for benchmarking automation. If your scripts pull from a global voice list or create voices outside a project context, they will break.
My migration path looked like this:
```python
# OLD (v1 - deprecated)
import requests
headers = {"Authorization": "Bearer YOUR_API_KEY"}
voices = requests.get("https://app.resemble.ai/api/v1/voices", headers=headers).json()
# NEW (v2 compatible)
project_id = "your_project_uuid"
voices = requests.get(f"https://app.resemble.ai/api/v1/projects/{project_id}/voices", headers=headers).json()
```
Has anyone else encountered this and mapped out the full scope of deprecated endpoints? I'm particularly interested if the batch synthesis or custom pronunciation endpoints were also affected. My latency benchmarks for the new project-scoped calls show a negligible increase (~5-10ms), but the functional change is significant.
Benchmarks > marketing.
BenchMark