Model reference · open weights

wav2vec2-ksponspeech

Available as managed deployment Audio Taeham · community Speech→text 1 variants 1k dl/mo

wav2vec2-ksponspeech is an open-weight audio or speech model from Taeham. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.

Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.

What it is

Released byTaeham
TypeAudio & music
TaskSpeech→text
Runs withtransformers
Released2022-06-11
Popularity1k downloads / month
LicenceOpen weights

About

What wav2vec2-ksponspeech is

This model is a fine-tuned version of Wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

  • WER(Word Error Rate) for Third party test data : 0.373

For improving WER:

  • Numeric / Character Unification
  • Decoding the word with the correct notation (from word based on pronounciation)
  • Uniform use of special characters (. / ?)
  • Converting non-existent words to existing words
Read the full model card

Model description

Korean Wav2vec with Ksponspeech dataset.

This model was trained by two dataset :

  • Train1 : https://huggingface.co/datasets/Taeham/wav2vec2-ksponspeech-train (1 ~ 20000th data in Ksponspeech)
  • Train2 : https://huggingface.co/datasets/Taeham/wav2vec2-ksponspeech-train2 (20100 ~ 40100th data in Ksponspeech)
  • Validation : https://huggingface.co/datasets/Taeham/wav2vec2-ksponspeech-test (20000 ~ 20100th data in Ksponspeech)
  • Third party test : https://huggingface.co/datasets/Taeham/wav2vec2-ksponspeech-test (60000 ~ 20100th data in Ksponspeech)

Hardward Specification

  • GPU : GEFORCE RTX 3080ti 12GB
  • CPU : Intel i9-12900k
  • RAM : 32GB

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0003
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 30
  • mixed_precision_training: Native AMP

Framework versions

  • Transformers 4.19.4
  • Pytorch 1.11.0
  • Datasets 2.2.2
  • Tokenizers 0.12.1

From the published model card. Full card on the HuggingFace links in the sidebar.

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys wav2vec2-ksponspeech for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (wav2vec2-ksponspeech below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -F model="wav2vec2-ksponspeech" -F file=@audio.mp3

Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms