ufakzeka-kulak: Building and Evaluating a 31.6M-Parameter Turkish Speech Recognizer from Scratch
cs.CL, eess.AS
Submitted: 2026-11-02
Updated: 2026-11-02
Comments: 9 pages (text on pages 1 to 8, references on pages 8 and 9). Model, card, attribution, code and demo: https://huggingface.co/ufakai/ufakzeka-kulak, https://github.com/ufakai/ufakzeka-kulak, https://kulak.ufakzeka.com
Code: https://github.com/ufakai/ufakzeka-kulak
License: http://creativecommons.org/licenses/by/4.0/
The gist: ufakzeka-kulak is an open Turkish speech-to-text model of 31.6M parameters, a FastConformer encoder with a token-and-duration transducer (TDT) and a CTC head…
Terminology
Abstract
ufakzeka-kulak is an open Turkish speech-to-text model of 31.6M parameters, a FastConformer encoder with a token-and-duration transducer (TDT) and a CTC head, trained from scratch and released under Apache-2.0. It writes cased, punctuated Turkish with numbers as words. Its 4,977 training hours come from sources whose licences allow commercial use; 281 hours are human-transcribed, and for the other 4,696 the training text is an uploader subtitle or a machine transcript, kept only where a second, independent reading of the audio agrees with it. The recipe, checkpoint, labelling and export choices were made on development data, and the test sets were scored once, in a release gate fixed before training. With greedy TDT decoding the normalised word error rate (WER) is 0.0921 (95 percent interval 0.0792 to 0.1055) on FLEURS Turkish test and 0.0916 (0.0873 to 0.0958) on Common Voice 27.0 Turkish test, lower than Whisper base and Whisper small on both sets with paired intervals wholly below zero; Whisper large-v3 (greedy) scores 0.0604 and 0.0767. On conversational speech the WER is 0.3807. We also report punctuation and casing, degraded audio, and how closely each shipped export reproduces the trained model.
Sources
- ufakzeka-1: Building and Evaluating a 151M-Parameter Turkish Language Model from Scratch
- ufakzeka-karar: An Open Turkish Typed-Decision Model with Order-Invariant Option Scoring
- Robust Speech Recognition via Large-Scale Weak Supervision
- OWSM v4: Improving Open Whisper-Style Speech Models via Data Scaling and Cleaning
- Omnilingual ASR: Open-Source Multilingual Speech Recognition for 1600+ Languages
- Qwen3-ASR Technical Report
- Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition
- Efficient Sequence Transduction by Jointly Predicting Tokens and Durations
- Intermediate Loss Regularization for CTC-based Speech Recognition
- CR-CTC: Consistency regularization on CTC for improved speech recognition
- BPE-Dropout: Simple and Effective Subword Regularization
- SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing
- SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition
- Attention Is All You Need
- Decoupled Weight Decay Regularization
- Scaling Laws and Compute-Optimal Training Beyond Fixed Training Durations
- MUSAN: A Music, Speech, and Noise Corpus
- Common Voice: A Massively-Multilingual Speech Corpus
- FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech
- YODAS: Youtube-Oriented Dataset for Audio and Speech
Related papers
- ALARM: Audio-Language Alignment for Reasoning Models
- ThinkSum: Probabilistic reasoning over sets using large language models
- Coherence boosting: When your pretrained language model is not paying enough attention
- From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for Role-Playing Agents
- Improving General Role-Playing Agents via Psychology-Grounded Reasoning and Role-Aware Policy Optimization
- The Art of Asking: Multilingual Prompt Optimization for Synthetic Data