--- language: es datasets: - ciempiess/ciempiess_light - ciempiess/ciempiess_balance - ciempiess/ciempiess_fem - common_voice - hub4ne_es_LDC98S74 - callhome_es_LDC96S35 tags: - audio - automatic-speech-recognition - spanish - xlrs-53-spanish - ciempiess - cimpiess-unam license: cc-by-4.0 model-index: - name: wav2vec2-large-xlsr-53-spanish-ep5-944h results: - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: Mozilla Common Voice 10.0 (Test) type: mozilla-foundation/common_voice_10_0 split: test args: language: es metrics: - name: WER type: wer value: 9.20 - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: Mozilla Common Voice 10.0 (Dev) type: mozilla-foundation/common_voice_10_0 split: validation args: language: es metrics: - name: WER type: wer value: 8.02 - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: CIEMPIESS-TEST type: ciempiess/ciempiess_test split: test args: language: es metrics: - name: WER type: wer value: 11.17 - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: 1997 Spanish Broadcast News Speech (HUB4-NE) type: HUB4NE_LDC98S74 split: test args: language: es metrics: - name: WER type: wer value: 7.48 - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: CALLHOME Spanish Speech (Test) type: callhome_LDC96S35 split: test args: language: es metrics: - name: WER type: wer value: 39.12 - task: name: Automatic Speech Recognition type: automatic-speech-recognition dataset: name: CALLHOME Spanish Speech (Dev) type: callhome_LDC96S35 split: validation args: language: es metrics: - name: WER type: wer value: 40.39 --- # wav2vec2-large-xlsr-53-spanish-ep5-944h **Paper:** [The state of end-to-end systems for Mexican Spanish speech recognition](http://journal.sepln.org/sepln/ojs/ojs/index.php/pln/article/viewFile/6485/3892) The "wav2vec2-large-xlsr-53-spanish-ep5-944h" is an acoustic model suitable for Automatic Speech Recognition in Spanish. It is the result of fine-tuning the model "facebook/wav2vec2-large-xlsr-53" for 5 epochs with around 944 hours of Spanish data gathered or developed by the [CIEMPIESS-UNAM Project](https://huggingface.co/ciempiess) since 2012. Most of the data is available at the the CIEMPIESS-UNAM Project homepage http://www.ciempiess.org/. The rest can be found in public repositories such as [LDC](https://www.ldc.upenn.edu/) or [OpenSLR](https://openslr.org/) The specific list of corpora used to fine-tune the model is: - [CIEMPIESS-LIGHT (18h25m)](https://catalog.ldc.upenn.edu/LDC2017S23) - [CIEMPIESS-BALANCE (18h20m)](https://catalog.ldc.upenn.edu/LDC2018S11) - [CIEMPIESS-FEM (13h54m)](https://catalog.ldc.upenn.edu/LDC2019S07) - [CHM150 (1h38m)](https://catalog.ldc.upenn.edu/LDC2016S04) - [TEDX_SPANISH (24h29m)](https://openslr.org/67/) - [LIBRIVOX_SPANISH (73h01m)](https://catalog.ldc.upenn.edu/LDC2020S01) - [WIKIPEDIA_SPANISH (25h37m)](https://catalog.ldc.upenn.edu/LDC2021S07) - [VOXFORGE_SPANISH (49h42m)](http://www.voxforge.org/es) - [MOZILLA COMMON VOICE 10.0 (320h22m)](https://commonvoice.mozilla.org/es) - [HEROICO (16h33m)](https://catalog.ldc.upenn.edu/LDC2006S37) - [LATINO-40 (6h48m)](https://catalog.ldc.upenn.edu/LDC95S28) - [CALLHOME_SPANISH (13h22m)](https://catalog.ldc.upenn.edu/LDC96S35) - [HUB4NE_SPANISH (31h41m)](https://catalog.ldc.upenn.edu/LDC98S74) - [FISHER_SPANISH (127h22m)](https://catalog.ldc.upenn.edu/LDC2010S01) - [Chilean Spanish speech data set (7h08m)](https://openslr.org/71/) - [Colombian Spanish speech data set (7h34m)](https://openslr.org/72/) - [Peruvian Spanish speech data set (9h13m)](https://openslr.org/73/) - [Argentinian Spanish speech data set (8h01m)](https://openslr.org/61/) - [Puerto Rico Spanish speech data set (1h00m)](https://openslr.org/74/) - [MediaSpeech Spanish (10h00m)](https://openslr.org/108/) - [DIMEX100-LIGHT (6h09m)](https://turing.iimas.unam.mx/~luis/DIME/CORPUS-DIMEX.html) - [DIMEX100-NIÑOS (08h09m)](https://turing.iimas.unam.mx/~luis/DIME/CORPUS-DIMEX.html) - [GOLEM-UNIVERSUM (00h10m)](https://turing.iimas.unam.mx/~luis/DIME/CORPUS-DIMEX.html) - [GLISSANDO (6h40m)](https://glissando.labfon.uned.es/es) - TELE_con_CIENCIA (28h16m) **Unplished Material** - UNSHAREABLE MATERIAL (118h22m) **Not available for sharing** The fine-tuning process was performed during November (2022) in the servers of the Language and Voice Lab (https://lvl.ru.is/) at Reykjavík University (Iceland) by Carlos Daniel Hernández Mena. # Evaluation ```python import torch from transformers import Wav2Vec2Processor from transformers import Wav2Vec2ForCTC #Load the processor and model. MODEL_NAME="carlosdanielhernandezmena/wav2vec2-large-xlsr-53-spanish-ep5-944h" processor = Wav2Vec2Processor.from_pretrained(MODEL_NAME) model = Wav2Vec2ForCTC.from_pretrained(MODEL_NAME) #Load the dataset from datasets import load_dataset, load_metric, Audio ds=load_dataset("ciempiess/ciempiess_test", split="test") #Downsample to 16kHz ds = ds.cast_column("audio", Audio(sampling_rate=16_000)) #Process the dataset def prepare_dataset(batch): audio = batch["audio"] #Batched output is "un-batched" to ensure mapping is correct batch["input_values"] = processor(audio["array"], sampling_rate=audio["sampling_rate"]).input_values[0] with processor.as_target_processor(): batch["labels"] = processor(batch["normalized_text"]).input_ids return batch ds = ds.map(prepare_dataset, remove_columns=ds.column_names,num_proc=1) #Define the evaluation metric import numpy as np wer_metric = load_metric("wer") def compute_metrics(pred): pred_logits = pred.predictions pred_ids = np.argmax(pred_logits, axis=-1) pred.label_ids[pred.label_ids == -100] = processor.tokenizer.pad_token_id pred_str = processor.batch_decode(pred_ids) #We do not want to group tokens when computing the metrics label_str = processor.batch_decode(pred.label_ids, group_tokens=False) wer = wer_metric.compute(predictions=pred_str, references=label_str) return {"wer": wer} #Do the evaluation (with batch_size=1) model = model.to(torch.device("cuda")) def map_to_result(batch): with torch.no_grad(): input_values = torch.tensor(batch["input_values"], device="cuda").unsqueeze(0) logits = model(input_values).logits pred_ids = torch.argmax(logits, dim=-1) batch["pred_str"] = processor.batch_decode(pred_ids)[0] batch["sentence"] = processor.decode(batch["labels"], group_tokens=False) return batch results = ds.map(map_to_result,remove_columns=ds.column_names) #Compute the overall WER now. print("Test WER: {:.3f}".format(wer_metric.compute(predictions=results["pred_str"], references=results["sentence"]))) ``` **Test Result**: 0.112 # BibTeX entry and citation info *When publishing results based on these models please refer to:* ```bibtex @misc{mena2022xlrs53spanish, title={Acoustic Model in Spanish: wav2vec2-large-xlsr-53-spanish-ep5-944h.}, author={Hernandez Mena, Carlos Daniel}, url={https://huggingface.co/carlosdanielhernandezmena/wav2vec2-large-xlsr-53-spanish-ep5-944h}, year={2022} } ``` # Acknowledgements The author wants to thank to the social service program ["Desarrollo de Tecnologías del Habla"](http://profesores.fi-b.unam.mx/carlos_mena/servicio.html) at the [Facultad de Ingeniería (FI)](https://www.ingenieria.unam.mx/) of the [Universidad Nacional Autónoma de México (UNAM)](https://www.unam.mx/). He also thanks to the social service students for all the hard work. Special thanks to Jón Guðnason, head of the Language and Voice Lab for providing computational power to make this model possible. The author also thanks to the "Language Technology Programme for Icelandic 2019-2023" which is managed and coordinated by Almannarómur, and it is funded by the Icelandic Ministry of Education, Science and Culture.