# WhisperBenchmarks

This repository provides easy-to-use benchmarks using audio and video content from the Internet Archive, specifically targeting various challenging scenarios in audio recordings.

Based on https://gitlab.com/aadnk/whisper-webui

## Models

| Model | Command |
|-|-|
|faster-large-v3| --whisper_implementation faster-whisper --model large-v3 |
|faster-medium| --whisper_implementation faster-whisper --model medium |
|faster-small| --whisper_implementation faster-whisper --model small |
|faster-tiny| --whisper_implementation faster-whisper --model tiny |

## Videos

Videos are chosen for being short and matching their given category

| Categories | Title | Link | Length | Instant Download |
|-|-|-|-|-|
| Poor mic placement | Body camera footage from July 10 traffic stop | [Internet Archive](https://archive.org/details/cobmn-Body_camera_footage_from_July_10_traffic_stop) | 2:22 | [MP4](https://archive.org/download/cobmn-Body_camera_footage_from_July_10_traffic_stop/Body_camera_footage_from_July_10_traffic_stop.mp4) |
| Thick accents | Moonshine for Medicine Popcorn Sutton | [Internet Archive](https://archive.org/details/this-is-the-last-dam-run-of-likker-ill-ever-make-full-movie/+Moonshine+for+Medicine++++Popcorn+Sutton.mp4) | 1:35 | [MP4](https://archive.org/download/this-is-the-last-dam-run-of-likker-ill-ever-make-full-movie/%20Moonshine%20for%20Medicine%20%20%20%20Popcorn%20Sutton.mp4) |
| Artifacts in audio | 2002 007 Movie Trailer Commercial Bad Video | [Internet Archive](https://archive.org/details/2002variouscommercials/2002+007+Movie+Trailer+Commercial+Bad+Video.mp4) | 0:14 | [MP4](https://archive.org/download/2002variouscommercials/2002%20A%20Touch%20Of%20Class%20Limos%20Bridal%20Show%20Wilton%20Mall%20Saratoga%20Commercial.mp4) |
| Ideal audio (one speaker) | 8 Bit Bookclub | [Internet Archive](https://archive.org/details/8-bit-bookclub/36+-+ANNOUNCEMENT++SUMMER+HIATUS.mp3) | 1:44 | [MP3](https://archive.org/download/8-bit-bookclub/36%20-%20ANNOUNCEMENT%20%20SUMMER%20HIATUS.mp3) |
| Long form (many speakers) | Bionic Woman "Black Magic" (1976) | [Internet Archive](https://archive.org/details/bionic-woman-black-magic) | 43:53 | [MP4](https://archive.org/download/bionic-woman-black-magic/Black%20Magic-NA_x264.mp4) |

## How to Run Whisper Benchmarks

-- TODO --

## Results

Results are for the complete run which includes loading the model, running VAD, and running the transcription. Links are embeded in the results for each category

### CPU Benchmarks

|  CPU Model | Poor mic placement (m:s:ms) | Thick accents (m:s:ms) | Artifacts in audio (m:s:ms) | Ideal audio (m:s:ms) | Long form | (Docker/Native) | Model |
|-|-|-|-|-|-|-|-|

### GPU Benchmarks

|  GPU Model | Poor mic placement (m:s:ms) | Thick accents (m:s:ms) | Artifacts in audio (m:s:ms) | Ideal audio (m:s:ms) | Long form (m:s:ms) | (Docker/Native) | Model |
|-|-|-|-|-|-|-|-|
| RTX 2060S | [00:11:12](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/2060S/Body_camera_footage_from_July_10_traffic_stop.mp4-subs.srt) | [00:06:96](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/2060S/Moonshine%20for%20Medicine%20Popcorn%20Sutton.mp4-subs.srt) | [00:04:40](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/2060S/2002%20007%20Movie%20Trailer%20Commercial%20Bad%20Video.mp4-subs.srt) | [00:08:31](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/2060S/36%20-%20ANNOUNCEMENT%20SUMMER%20HIATUS.mp3-subs.srt) | [03:34:00](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/2060S/Black%20Magic-NA_x264.mp4-subs.srt) | Native | faster-medium |
| RTX A5000 | [00:10:86](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/A5000/Body_camera_footage_from_July_10_traffic_stop.mp4-subs.srt) | [00:07.84](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/A5000/Moonshine%20for%20Medicine%20Popcorn%20Sutton.mp4-subs.srt) | [00:07:75](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/A5000/2002%20007%20Movie%20Trailer%20Commercial%20Bad%20Video.mp4-subs.srt) | [00:08:91](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/A5000/36%20-%20ANNOUNCEMENT%20SUMMER%20HIATUS.mp3-subs.srt) | [03:17:10](https://git.myco.systems/brooke/whisperbenchmarks/src/branch/main/benchmark-outputs/A5000/Black%20Magic-NA_x264.mp4-subs.srt) | Native | faster-large-v3 |

## Todo:
- [ ] Write easy bash scripts for running a set of benchmarks with an easy cleanup
- [ ] Finalize a standard format for exporting the data into a spreadsheet
- [ ] Some outputs seem to be presenting artifacts when using faster-whisper, there should be some kind of easy review process so that batches can quickly be re-run, maybe a link to open vlc with the new subtitles loaded?

## Notes:
From running multiple benchmarks the resulting time it takes seems to vary a lot more than I would have expected for shorter clips, "Poor Mic placement" tested anywhere from 8s to 20s on an A5000.