---
title: tujia-xlsr53-ft
canonical_url: "https://www.modelscope.cn/models/TujiaNLP/tujia-xlsr53-ft"
md_url: "https://www.modelscope.cn/models/TujiaNLP/tujia-xlsr53-ft.md"
repository: TujiaNLP/tujia-xlsr53-ft
last_updated: 2026-10-28
license: other
pipeline_tag: feature-extraction
tasks:
  - feature-extraction
model_type:
  - wav2vec2
architectures:
  - Wav2Vec2Model
parameters: 315.4M
tensor_type:
  - F32
library_name:
  - transformer
  - safetensors
language:
  - tji
downloads: 10
stars: 0
tags:
  - wav2vec2
  - speech
  - tujia
  - tji
  - encoder
---

# tujia-xlsr53-ft

> tujia-xlsr53-ft - TujiaNLP 在 ModelScope 开源的模型。Fine-tuned XLSR-53 encoder for Tujia (Northern dialect). This is only the encoder; it is not a complete ASR model and produces no transcript on its own. It is the feature extractor required by the tujia-ctc v0.2…

TujiaNLP/tujia-xlsr53-ft 是 ModelScope 魔搭社区上的 315.4M 参数feature-extraction模型，采用 other 许可。

- **Repository**: TujiaNLP/tujia-xlsr53-ft
- **License**: other
- **Tasks**: feature-extraction
- **Parameters**: 315.4M
- **Tags**: wav2vec2, speech, tujia, tji, encoder
- **Downloads**: 10
- **Stars**: 0
- **Last updated**: 2026-10-28

Source: https://www.modelscope.cn/models/TujiaNLP/tujia-xlsr53-ft

---

# tujia-xlsr53-ft

Fine-tuned XLSR-53 encoder for Tujia (Northern dialect). This is **only the encoder**;
it is not a complete ASR model and produces no transcript on its own. It is the feature
extractor required by the `tujia-ctc v0.2` syllable recognizer.

- Base model: `facebook/wav2vec2-xlsr-53-espeak-cv-ft` (Apache-2.0)
- Fine-tuning: the top 4 of 12 transformer layers were trainable (`freeze_layers = 8`)
  on ~2.6 hours of Tujia read speech, jointly with a 709-class toned-syllable CTC head
- Size: 315,438,720 parameters, fp32, ~1.26 GB across 5 standard safetensors shards
- Output: 1024-dim frame features at 50 Hz, identical interface to the base model

## Do not substitute the original XLSR-53

The features of this encoder differ from `facebook/wav2vec2-xlsr-53-espeak-cv-ft`.
Using the original weights with the `tujia-ctc v0.2` head degrades recognition
substantially; the head was trained against these features.

## Usage

```python
from transformers import Wav2Vec2FeatureExtractor, Wav2Vec2Model

feat = Wav2Vec2FeatureExtractor.from_pretrained("xbaibing/tujia-xlsr53-ft")
enc = Wav2Vec2Model.from_pretrained("xbaibing/tujia-xlsr53-ft")
```

The v0.2 release archive fetches this repository through `fetch_encoder.sh`, which
places it in `models/tujia-xlsr53-ft` and points `TUJIA_XLSR` at it.

## Data and license

- Upstream weights: Apache-2.0, attribution must be retained.
- These fine-tuned weights: research use only. No rights to the training recordings or
  transcripts are granted, and neither is distributed.
- Parent project: https://github.com/ibaibing/TujiaNLP
- Model page: https://github.com/ibaibing/tujia-asr