Implementation of Native BWF (Broadcast Wave Format) Metadata Preservation

The Problem:
In professional audio engineering, sound restoration, and cinematic post-production workflows, keeping exact timeline synchronization is mandatory. The industry-standard protocol for this is BWF (Broadcast Wave Format), which embeds a precise TimeReference (the absolute sample-count timestamp since the project's timeline zero-point) inside the WAV container's metadata chunk (bext).

The process is roughly as follows:
In professional digital audio workstations (DAWs, such as REAPER, Pro Tools, or Cubase), fragments containing defects are marked in a track to be restored or regenerated based on a trained voice model, creating patches. These fragments are exported with BWF tags, which also contain a TimeReference. After converting the source files to RVC, the TimeReference must be manually copied from the source files to the corresponding converted files (for example, in program BWF MetaEdit).

Currently, when any audio chunk is processed through either the RVC Single Inference or Batch Inference pipelines, the underlying audio-writing libraries (such as scipy.io.wavfile or soundfile) re-synthesize the output PCM array from scratch as a raw structural matrix. During this generation stage, all original structural headers and bext meta-tags are completely purged.

Suggested Solution:
Implement a native metadata extraction and inheritance safety protocol directly inside the inference core pipeline scripts (infer/vc/modules.py and infer/vc/pipeline.py). Before the generator executes and renders the final target audio arrays, the script should scan the primary source input file for the presence of a valid BWF header. If detected, the core should preserve the original bext block context and automatically re-compile/inject those exact metadata parameters into the newly synthesized target .wav container upon saving.

Implementing this standard 4-line metadata inheritance pass will instantly bridge the gap between AI synthesis models and industrial-grade professional audio engineering ecosystems, unlocking highly automated multi-patch non-destructive workflow configurations for thousands of studio engineers.

Start in infer/vc/modules.py and infer/vc/pipeline.py, tracing the Single Inference and Batch Inference paths from source input to the saved WAV output. Inspect how the current audio-writing libraries create files and how a source BWF bext block could be detected and carried forward. Done means converted outputs retain the source BWF metadata when present without breaking ordinary WAV processing.