Fix Apple Silicon source install requirements

[yyovil]

Why this helps

The README currently promises that source users can set up the project with pip install -r requirements.txt.

Before this PR, that promise was broken on Apple Silicon and other non-CUDA environments because the requirements file forced a CUDA-specific PyTorch build and pinned kaldi_native_fbank to a version without a native macOS arm64 wheel.

After this PR, the source install path becomes platform-neutral again: pip can choose a compatible PyTorch wheel for the current environment, and it can resolve a newer kaldi_native_fbank build with Apple Silicon support.

Before

• pip install -r requirements.txt failed on native Apple Silicon environments
• torch==2.1.0+cu118 forced a CUDA-only wheel in a general-purpose requirements file
• kaldi_native_fbank==1.15 blocked pip from selecting newer arm64 macOS wheels

After

• pip install -r requirements.txt resolves on Apple Silicon native Python
• source installs no longer assume a CUDA-specific PyTorch build
• CUDA users still have a clear path by installing their preferred PyTorch wheel separately
• the documented source install flow better matches what the repo promises

Changes

• replace the CUDA-specific Torch requirement in requirements.txt with a platform-neutral PyTorch dependency
• relax the kaldi_native_fbank pin so Apple Silicon can resolve a compatible wheel
• document the expected PyTorch/CUDA behavior in the README

Testing

• python -m pip install --dry-run --ignore-installed --no-deps -r requirements.txt
• python -m pip install --dry-run --ignore-installed --no-deps --only-binary=:all: --platform macosx_11_0_arm64 --python-version 3.11 --implementation cp --abi cp311 -r requirements.txt

Closes #8.

[Copilot]

Pull request overview

This PR updates the source-install dependency set to better support Apple Silicon and other CPU-only environments by removing CUDA-specific PyTorch constraints and loosening a problematic wheel pin, with a README note for CUDA users.

Changes:

• Switch requirements.txt from a CUDA-suffixed Torch pin to a platform-neutral PyTorch spec.
• Relax the kaldi_native_fbank version constraint to allow resolving Apple Silicon-compatible wheels.
• Document in the README that CUDA users should install their preferred PyTorch build separately for source installs.

Reviewed changes

Copilot reviewed 1 out of 2 changed files in this pull request and generated no comments.

File	Description
requirements.txt	Replaces CUDA-specific Torch pin with a generic Torch constraint; loosens kaldi_native_fbank pin.
README.md	Adds guidance clarifying PyTorch installation expectations for CUDA users on the source-install path.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.