Serkali-sudo's picture
Upload validated SpeechX LiteRT fp16 export
11b11bb verified
|
Raw History Blame Contribute Delete
1.09 kB
---
license: apache-2.0
base_model: AhmedReda25300/SpeechX
pipeline_tag: audio-classification
tags:
- litert
- tflite
---
# AutoSub Voice Classifier for LiteRT
A LiteRT conversion of [SpeechX by AhmedReda25300](https://huggingface.co/AhmedReda25300/SpeechX),
which is based on the [OpenAI Whisper Base encoder](https://huggingface.co/openai/whisper-base).
The export keeps SpeechX's two-class voice classifier (`female` and `male`)
for suggesting dubbing voices in AutoSub. It omits the language-identification
head. The output describes the model's classification of the audio.
## Export
- `speechx-fp16.tflite` — LiteRT CPU graph with FP16 weight storage.
- `mel_filters_f32le.bin` — Mel filters for 16 kHz mono audio preprocessing.
- `speechx.json` — Tensor shapes, label order, source revision, and file checksums.
- `validation.json` — CPU comparison against the original SpeechX classifier on
the test clips. This measures conversion parity.
## License
SpeechX is released under Apache 2.0. See [LICENSE](LICENSE) and
[OpenAI Whisper's MIT notice](LICENSE-whisper-MIT).