README.md

---
language: sv
license: apache-2.0
tags:
- seq-2-seq asr
- whisper
datasets:
- mozilla-foundation/common_voice_11_0
library_name: torch
---

## Model description

This model is a fine-tuned version of OpenAI's pre-trained Whisper small. [Source](https://github.com/openai/whisper)

The model were trained on 680 000 hours of audio with corresponding transcripts from the internet, 65% of which was in english audio and 83 % of which had english transcripts.

The model was then further trained for 4000 iterations, 500 of which as warm-up, on Swedish data from [Common_voice 11.0](https://huggingface.co/datasets/mozilla-foundation/common_voice_11_0). Achieving a WER of 19.865.


 ## Training Metrics
![Training table](whisper_finetune.png)
 ## Model Plot

<details>
<summary>View Training Plots</summary>

![Metrics](whisper_metrics.png)

</details>