Move torchcodec to optional dependencies

[onuryuruten]

Description

Moves torchcodec from required dependencies to optional dependencies to improve installation reliability and reduce unnecessary requirements for core TTS functionality.

Problem

Currently, torchcodec is listed as a required dependency, but:

1. Not needed for core TTS functionality - F5-TTS works perfectly fine without it for text-to-speech tasks
2. Causes installation failures - Users encounter errors like AttributeError: module 'torchcodec' has no attribute 'decoders' due to incomplete or broken installations, particularly on macOS Intel machines
3. Platform compatibility issues - torchcodec has varying levels of support across different platforms and architectures
4. Increases installation complexity - Adds an unnecessary dependency that most users won't utilize

Use Case Analysis

torchcodec is primarily useful for:

• Video decoding and processing
• Extracting audio from video files
• Multimodal workflows

For standard text-to-speech workflows (which represent the majority of F5-TTS use cases), torchcodec is never utilized.

Solution

Move torchcodec to optional dependencies under a new video extra (or include in full extra if preferred).

Changes

# Before
dependencies = [
    ...
    "torchcodec",
    ...
]

# After
dependencies = [
    ...
    # torchcodec moved to optional dependencies
    ...
]

[project.optional-dependencies]
video = [
    "torchcodec",
]

Users who need video processing capabilities can install with:

pip install f5-tts[video]

Benefits

• ✅ Reduces installation failures for standard users
• ✅ Maintains functionality for users who need video processing
• ✅ Improves cross-platform compatibility
• ✅ Follows Python packaging best practices (keep required deps minimal)
• ✅ Faster installation times for most users

Testing

• Verified F5-TTS runs successfully without torchcodec for standard TTS workflows
• No breaking changes for users who install with [video] extra

Migration Path

Users currently relying on torchcodec (if any) can simply install with:

pip install f5-tts[video]

Or add it manually:

pip install torchcodec

Alternative Consideration

If the maintainers prefer, torchcodec could be added to an existing full extra instead of creating a new video extra.