feat: upgrade OpenAudio backend to Fish Speech S2-Pro and refactor de…#29
feat: upgrade OpenAudio backend to Fish Speech S2-Pro and refactor de…#29KevinBonnoron wants to merge 1 commit intomainfrom
Conversation
WalkthroughThe PR replaces the OpenAudio TTS backend with a new Fish Audio backend implementation. Changes include: removing the openaudio backend class and installation function while adding a FishAudioBackend class with streaming support; updating the dependency registry to use "fish_audio" key with pinned torch/torchaudio versions (2.8.0) and installing fish-speech from v2.0.0-beta git revision; removing OpenAudio S1 Mini model from the catalog and adding Fish Speech S2-Pro model; and updating the backend registry imports to reference the new backend. Possibly related issues
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. 📝 Coding Plan
Comment Tip CodeRabbit can use oxc to improve the quality of JavaScript and TypeScript code reviews.Add a configuration file to your project to customize how CodeRabbit runs oxc. |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: ac30a16c-ddef-4222-8c79-71cbd6762ee4
📒 Files selected for processing (5)
inference/src/backends/deps.pyinference/src/backends/fish_audio.pyinference/src/backends/openaudio.pyinference/src/backends/registry.pyserver/src/manifest/models.manifest.ts
💤 Files with no reviewable changes (1)
- inference/src/backends/openaudio.py
|
|
||
| _TORCH_CPU_INDEX = "https://download.pytorch.org/whl/cpu" | ||
| _TORCH = ["torch>=2.5.0,<2.8", "torchaudio>=2.5.0,<2.8"] | ||
| _TORCH_28 = ["torch==2.8.0", "torchaudio==2.8.0"] |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Check if fish-speech v2.0.0-beta actually requires torch 2.8.0
curl -s "https://raw.githubusercontent.com/fishaudio/fish-speech/v2.0.0-beta/pyproject.toml" 2>/dev/null | grep -A 10 "dependencies" || \
curl -s "https://raw.githubusercontent.com/fishaudio/fish-speech/v2.0.0-beta/setup.py" 2>/dev/null | grep -A 10 "install_requires"Repository: KevinBonnoron/sirene
Length of output: 467
🏁 Script executed:
cat -n inference/src/backends/deps.py | head -100Repository: KevinBonnoron/sirene
Length of output: 3904
🏁 Script executed:
cat -n inference/src/backends/deps.py | tail -100Repository: KevinBonnoron/sirene
Length of output: 4333
🏁 Script executed:
# Search for how install_backend_deps is called
rg "install_backend_deps" --type py -B 2 -A 2Repository: KevinBonnoron/sirene
Length of output: 2877
🏁 Script executed:
# Search for README or documentation about backends
fd -e md -e txt | grep -i readmeRepository: KevinBonnoron/sirene
Length of output: 73
🏁 Script executed:
head -150 README.mdRepository: KevinBonnoron/sirene
Length of output: 2720
PyTorch version constraint conflict with other backends.
The _TORCH_28 constant pins torch==2.8.0, which violates the torch>=2.5.0,<2.8 constraint used by other backends (qwen, f5-tts, cosyvoice, chatterbox, higgs_audio). Installing multiple backends in the same environment will fail with a pip resolver error since torch==2.8.0 does not satisfy <2.8.
Either document that fish_audio cannot coexist with other torch-based backends, or adjust the constraint to allow flexibility (e.g., torch>=2.8.0,<3.0 for all backends if compatible).
| packages=[ | ||
| *_TORCH_28, | ||
| "transformers<=4.57.3", |
There was a problem hiding this comment.
🧹 Nitpick | 🔵 Trivial
Unusual transformers version constraint.
The constraint transformers<=4.57.3 only sets an upper bound without a lower bound, which is atypical. Other backends in this file use >=4.47.0. This could allow installation of very old transformers versions. Consider adding a lower bound:
- "transformers<=4.57.3",
+ "transformers>=4.47.0,<=4.57.3",📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| packages=[ | |
| *_TORCH_28, | |
| "transformers<=4.57.3", | |
| packages=[ | |
| *_TORCH_28, | |
| "transformers>=4.47.0,<=4.57.3", |
| if not codes_list: | ||
| raise RuntimeError("OpenAudio LLM generated no tokens") |
There was a problem hiding this comment.
Fix incorrect backend name in error message.
The error message references "OpenAudio LLM" but this is the Fish Audio backend. This appears to be a copy-paste artifact from the removed OpenAudio implementation.
📝 Proposed fix
if not codes_list:
- raise RuntimeError("OpenAudio LLM generated no tokens")
+ raise RuntimeError("Fish Audio model generated no tokens")🧰 Tools
🪛 Ruff (0.15.6)
[warning] 107-107: Avoid specifying long messages outside the exception class
(TRY003)
1 participant
…pendencies
Summary by CodeRabbit
New Features
Removals