# romanian-wav2vec2 by Théo Gigant: Open-Weight Model Source: https://savrn.com/models/romanian-wav2vec2 Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve romanian-wav2vec2 (315M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them. | Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits | | --- | --- | --- | --- | --- | --- | | 16-bit | 0.6 GB | 0.8 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 8-bit | 0.3 GB | 0.4 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 4-bit | 0.2 GB | 0.2 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the [SAVRN Index](https://savrn.com/ai-index/pricing/gpus), read Oct 24, 2026. [romanian-wav2vec2 on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/romanian-wav2vec2/gpus) ## SAVRN's Notes on romanian-wav2vec2 Romanian audio in, Romanian text out, and almost nothing on the meter. At 16-bit the weights are 0.6 GB and the memory needed is 0.8 GB, under half of one percent of the 192 GB on the cheapest setup the Index prices for it, one MI300X at $1.85 an hour on demand. Nobody rents that card for one 315M-parameter recognizer; the question is how many streams share a GPU you already own. Apache 2.0 permits commercial use, modification and redistribution, so it can sit inside a paid product, provided you keep the notices and state your changes. Check the lineage: it descends from facebook/wav2vec2-xls-r-300m, trained on Common Voice 8.0 Romanian plus a Romanian speech synthesis set, last touched September 2023. Then decide whether the 5-gram language model ships with it, because the publisher's Common Voice test word error rate is 11.73 without it and 7.31 with it. SAVRN Research, 2026-09-18 ## Model Card By Théo Gigant, published under apache-2.0, revision 79cf603aac59. You can test this model online with the [Space for Romanian Speech Recognition](https://huggingface.co/spaces/gigant/romanian-speech-recognition) The model ranked TOP-1 on Romanian Speech Recognition during HuggingFace's Robust Speech Challenge : - [The Speech Bench](https://huggingface.co/spaces/huggingface/hf-speech-bench) - [Speech Challenge Leaderboard](https://huggingface.co/spaces/speech-recognition-community-v2/FinalLeaderboard) ### Romanian Wav2Vec2 This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the [Common Voice 8.0 - Romanian subset](https://huggingface.co/datasets/mozilla-foundation/common_voice_8_0) dataset, with extra training data from [Romanian Speech Synthesis](https://huggingface.co/datasets/gigant/romanian_speech_synthesis_0_8_1) dataset. Without the 5-gram Language Model optimization, it achieves the following results on the evaluation set (Common Voice 8.0, Romanian subset, test split): - Loss: 0.1553 - Wer: 0.1174 - Cer: 0.0294 ### Model description The architecture is based on [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) with a speech recognition CTC head and an added 5-gram language model (using [pyctcdecode](https://github.com/kensho-technologies/pyctcdecode) and [kenlm](https://github.com/kpu/kenlm)) trained on the Romanian Corpora Parliament dataset. Those libraries are needed in order for the language model-boosted decoder to work. ### Intended uses & limitations [Read the full model card (919 words)](https://savrn.com/models/romanian-wav2vec2/card) ## Configuration Architecture Wav2Vec2ForCTC Layers 24 Hidden size 1,024 Feed-forward size 4,096 Attention heads 16 Vocabulary size 36 Stored precision float32 Model type wav2vec2 ## Identity and Version Repository gigant/romanian-wav2vec2 Publisher Théo Gigant Task Speech recognition Modality Audio Library transformers Parameters 315M parameters Languages ro Revision 79cf603aac59501d02bfeb37f615efc3ac4ce1b3 First published 2022-03-02 Last updated 2023-09-13 ## Files and Weights 22 files, 3.0 GB in total. The weights are 4 files totalling 3.0 GB in bin, safetensors. Weights4 files · 3.0 GB Configuration11 files · 35.7 KB Tokenizer2 files · 639 B Documentation1 file · 11.2 KB Other2 files · 911.0 KB Repository2 files · 1.2 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | language_model/5gram.bin | Weights | 491.8 MB | b03301f34e51 | | model.safetensors | Weights | 1.3 GB | a42ddf6318df | | pytorch_model.bin | Weights | 1.3 GB | d0b801c7a4fa | | training_args.bin | Weights | 3.1 KB | cd7ffa627b29 | | added_tokens.json | Configuration | 23 B | — | | all_results.json | Configuration | 442 B | — | | alphabet.json | Configuration | 243 B | — | | config.json | Configuration | 2.0 KB | — | | eval.py | Configuration | 5.1 KB | — | | eval_results.json | Configuration | 265 B | — | | language_model/attrs.json | Configuration | 78 B | — | | preprocessor_config.json | Configuration | 260 B | — | | special_tokens_map.json | Configuration | 502 B | — | | train_results.json | Configuration | 198 B | — | | trainer_state.json | Configuration | 26.5 KB | — | | README.md | Documentation | 11.2 KB | — | | eval.sh | Other | 215 B | — | | language_model/unigrams.txt | Other | 910.8 KB | — | | .gitattributes | Repository | 1.2 KB | — | | .gitignore | Repository | 13 B | — | | tokenizer_config.json | Tokenizer | 330 B | — | | vocab.json | Tokenizer | 309 B | — | ## License and Download License apache-2.0 Access Open weights, no gate Download size 3.0 GB [Download from Théo Gigant](https://huggingface.co/gigant/romanian-wav2vec2) Released by Théo Gigant through its official repository on Hugging Face. [Read the license](https://www.apache.org/licenses/LICENSE-2.0). ## Built From - Derived from facebook/wav2vec2-xls-r-300m - Trained on (disclosed) gigant/romanian_speech_synthesis_0_8_1 - Trained on (disclosed) mozilla-foundation/common_voice_8_0 ## Evaluations Each result is shown as reported, with the conditions its reporter stated. None is a SAVRN measurement. A comparison lines two results up only when their configuration, unit and setup are all stated and identical. | Benchmark | Conditions | Result | Reported by | Revision | Date | | --- | --- | --- | --- | --- | --- | | Common Voice | Task Automatic Speech RecognitionMetric Test CER (with LM)Comparison conditions not established | 2.17 | [gigant](https://huggingface.co/gigant/romanian-wav2vec2) Publisher reported | Evaluated revision not stated | — | | Common Voice | Task Automatic Speech RecognitionMetric Test CER (without LM)Comparison conditions not established | 2.93 | [gigant](https://huggingface.co/gigant/romanian-wav2vec2) Publisher reported | Evaluated revision not stated | — | | Common Voice | Task Automatic Speech RecognitionMetric Test WER (with LM)Comparison conditions not established | 7.31 | [gigant](https://huggingface.co/gigant/romanian-wav2vec2) Publisher reported | Evaluated revision not stated | — | | Common Voice | Task Automatic Speech RecognitionMetric Test WER (without LM)Comparison conditions not established | 11.73 | [gigant](https://huggingface.co/gigant/romanian-wav2vec2) Publisher reported | Evaluated revision not stated | — | | Robust Speech Event | Task Automatic Speech RecognitionMetric Dev CER (with LM)Comparison conditions not established | 14.52 | [gigant](https://huggingface.co/gigant/romanian-wav2vec2) Publisher reported | Evaluated revision not stated | — | | Robust Speech Event | Task Automatic Speech RecognitionMetric Dev CER (without LM)Comparison conditions not established | 16.04 | [gigant](https://huggingface.co/gigant/romanian-wav2vec2) Publisher reported | Evaluated revision not stated | — | | Robust Speech Event | Task Automatic Speech RecognitionMetric Dev WER (with LM)Comparison conditions not established | 38.63 | [gigant](https://huggingface.co/gigant/romanian-wav2vec2) Publisher reported | Evaluated revision not stated | — | | Robust Speech Event | Task Automatic Speech RecognitionMetric Dev WER (without LM)Comparison conditions not established | 46.99 | [gigant](https://huggingface.co/gigant/romanian-wav2vec2) Publisher reported | Evaluated revision not stated | — | | Robust Speech Event - Test Data | Task Automatic Speech RecognitionMetric Test WERComparison conditions not established | 43.23 | [gigant](https://huggingface.co/gigant/romanian-wav2vec2) Publisher reported | Evaluated revision not stated | — | ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 3.0 GB | | 16-bit | 0.6 GB | | 8-bit | 0.3 GB | | 4-bit | 0.2 GB | Weights only, from the published parameter count; the key-value cache and runtime add to this. ## Questions About romanian-wav2vec2 ### How much GPU memory does romanian-wav2vec2 need? About 0.8 GB at 16-bit and 0.2 GB at 4-bit: the weights (315M parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run romanian-wav2vec2 on? At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists. ### Can I use romanian-wav2vec2 commercially? Yes. romanian-wav2vec2 is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors. ## Similar Models Model · Speech recognition ### [wav2vec2-xls-r-300m-ftspeech](https://savrn.com/models/wav2vec2-xls-r-300m-ftspeech) [Dan Saattrup Smart](https://savrn.com/model-publishers/saattrupdan) This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the FTSpeech dataset, being a dataset of 1,800 hours of transcribed speeches from the Danish parliament. The model achieves the following WER scores (lower is better): The use of this model needs to adhere to this license from the Danish Parliament. Open weights other 315M parameters transformers [View model](https://savrn.com/models/wav2vec2-xls-r-300m-ftspeech) Model · Speech recognition ### [wav2vec2-xls-r-300m-hebrew](https://savrn.com/models/wav2vec2-xls-r-300m-hebrew) [Vladimir Gurevich](https://savrn.com/model-publishers/imvladikon) This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the private datasets in 2 stages - firstly was fine-tuned on a small dataset with good samples Then the obtained model was fine-tuned on a large dataset with the small good dataset, with various samples from different sources, and with an unlabeled dataset that was weakly labeled using a previously trained model. (weakly labeled data wasn't used in validation set) on small dataset on large dataset on small dataset on large dataset The following hyperparameters were used during training: - learningrate: 0.0003 - trainbatchsize: 8 - evalbatchsize: 8 - distributedtype: multi-GPU - numdevices: 2 - gradientaccumulationsteps: 4… Open weights 315M parameters transformers [View model](https://savrn.com/models/wav2vec2-xls-r-300m-hebrew) Model · Speech recognition ### [w2v-xls-r-uk](https://savrn.com/models/w2v-xls-r-uk) [Smoliakov](https://savrn.com/model-publishers/yehor) ATTENTION! Metrics (float16) using evaluate library with batchsize=1 Open weights apache-2.0 315M parameters transformers [View model](https://savrn.com/models/w2v-xls-r-uk) Model · Speech recognition ### [mms-300m-1130-forced-aligner](https://savrn.com/models/mms-300m-1130-forced-aligner) [Mahmoud Ashraf](https://savrn.com/model-publishers/mahmoudashraf) This Python package provides an efficient way to perform forced alignment between text and audio using Hugging Face's pretrained models. it also features an improved implementation to use much less memory than TorchAudio forced alignment API. The model checkpoint uploaded here is a conversion from torchaudio to HF Transformers for the MMS-300M checkpoint trained on forced alignment dataset Open weights cc-by-nc-4.0 315M parameters transformers [View model](https://savrn.com/models/mms-300m-1130-forced-aligner) Model · Speech recognition ### [wav2vec2-large-voxrex-swedish](https://savrn.com/models/wav2vec2-large-voxrex-swedish) [National Library of Sweden / KBLab](https://savrn.com/model-publishers/kblab) Finetuned version of KBs VoxRex large model using Swedish radio broadcasts, NST and Common Voice data. Evalutation without a language model gives the following: WER for NST + Common Voice test set (2% of total sentences) is 2.5%. WER for Common Voice test set is 8.49% directly and 7.37% with a 4-gram language model. When using this model, make sure that your speech input is sampled at 16kHz. This model has been fine-tuned for 120000 updates on NST + CommonVoice and then for an additional 20000 updates on CommonVoice only. The additional fine-tuning on CommonVoice hurts performance on the NST+CommonVoice test set somewhat and, unsurprisingly, improves it on the CommonVoice test set. It seems… Open weights cc0-1.0 315M parameters transformers [View model](https://savrn.com/models/wav2vec2-large-voxrex-swedish) Model · Speech recognition ### [wav2vec2-xls-r-300m-cs-250](https://savrn.com/models/wav2vec2-xls-r-300m-cs-250) [Vojtěch Drábek](https://savrn.com/model-publishers/comodoro) This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the commonvoice 8.0 dataset as well as other datasets listed below. It achieves the following results on the evaluation set: The eval.py script results using a LM are: Fine-tuned facebook/wav2vec2-large-xlsr-53 on Czech using the Common Voice dataset. When using this model, make sure that your speech input is sampled at 16kHz. The model can be used directly (without a language model) as follows: The model can be evaluated using the attached eval.py script: The Common Voice 8.0 train and validation datasets were used for training, as well as the following datasets: - Šmídl, Luboš and Pražák, Aleš, 2013, OVM – Otázky… Open weights apache-2.0 315M parameters transformers [View model](https://savrn.com/models/wav2vec2-xls-r-300m-cs-250) ## Théo Gigant [All models and datasets](https://savrn.com/model-publishers/gigant) ## Versions - [79cf603aac59](https://savrn.com/models/romanian-wav2vec2/versions/79cf603aac59) · current 2026-10-24 ## Explore More - [All speech recognition models](https://savrn.com/models/tasks/speech-recognition) - [All models under apache-2.0](https://savrn.com/models/licenses/apache-2-0) - [Model comparisons](https://savrn.com/models/comparisons) - [The model directory](https://savrn.com/models) - [Open model prices by host](https://savrn.com/ai-index/pricing/open-models) ## Source - Repository metadata, read 2026-10-19. - [Hugging Face record](https://huggingface.co/gigant/romanian-wav2vec2) - [How the hub is built](https://savrn.com/model-hub/methodology)