GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
Sunbird
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
Sunbird
Model Tree
huwenjie333/whisper-v3-ft-af51
Input Modalities
Output Modalities
Supported Functionality
You can use this Colab Notebook to try out the model, where you can transcribe for an audio sample from Hugging Face dataset, for a uploaded audio file, or for your own recording from the computer microphone.
The model is used in a similar way as the base Whisper model. For better accuracy, please specify the language during generation.
See the example below, which transcribes for an audio sample from a Hugging Face dataset :
python
import transformersimport datasetsimport torchSAMPLE_RATE = 16000LANGUAGE_TOKENS_WHISPER = {# Existing languges codes from Whisper"eng": 50259, "fra": 50265, "swa": 50318, "sna": 50324, "yor": 50325, "som": 50326,"afr": 50327, "amh": 50334, "mlg": 50349, "lin": 50353, "hau": 50354,# Overwrite unused language tokens"ach": 50357, "aka": 50356, "bam": 50355, "bem": 50352, "ber": 50351,"cgg": 50350, "dag": 50348, "dga": 50347, "ewe": 50346, "ful": 50345,"ibo": 50344, "kab": 50343, "kau": 50342, "kik": 50341, "kin": 50340,"kln": 50339, "koo": 50338, "kpo": 50337, "led": 50336, "lgg": 50335,"lth": 50333, "lug": 50332, "luo": 50331, "luy": 50330, "myx": 50329,"nbl": 50328, "nya": 50323, "nyn": 50322, "orm": 50321, "pcm": 50320,"ruc": 50319, "rwm": 50317, "sot": 50316, "teo": 50315, "tsn": 50314,"ttj": 50313, "wol": 50312, "xho": 50311, "xog": 50310, "zul": 50309}LANGUAGE_NAMES = {'Acholi': 'ach', 'Afrikaans': 'afr', 'Akan': 'aka', 'Amharic': 'amh', 'Ateso': 'teo','Bambara': 'bam', 'Bemba': 'bem', 'Berber': 'ber', 'Chichewa': 'nya', 'Dagaare': 'dga','Dagbani': 'dag', 'English': 'eng', 'Ewe': 'ewe', 'French': 'fra', 'Fulani': 'ful','Hausa': 'hau', 'Igbo': 'ibo', 'Ikposo': 'kpo', 'Kabyle': 'kab', 'Kalenjin': 'kln','Kanuri': 'kau', 'Kikuyu': 'kik', 'Kinyarwanda': 'kin', 'Kwamba': 'rwm', 'Lendu': 'led','Lingala': 'lin', 'Luganda': 'lug', 'Lugbara': 'lgg', 'Luhya': 'luy', 'Lumasaba': 'myx','Luo': 'luo', 'Lusoga': 'xog', 'Malagasy': 'mlg', 'Ndebele': 'nbl', 'Nigerian Pidgin': 'pcm','Oromo': 'orm', 'Rukiga': 'cgg', 'Rukonjo': 'koo', 'Runyankole': 'nyn', 'Ruruuli': 'ruc','Rutooro': 'ttj', 'Shona': 'sna', 'Somali': 'som', 'Sotho': 'sot', 'Swahili': 'swa','Thur': 'lth', 'Tswana': 'tsn', 'Wolof': 'wol', 'Xhosa': 'xho', 'Yoruba': 'yor', 'Zulu': 'zul'}model_id = "Sunbird/asr-whisper-51-african-languages"model = transformers.WhisperForConditionalGeneration.from_pretrained(model_id)processor = transformers.WhisperProcessor.from_pretrained(model_id)def transcribe_by_whisper(audio_array, language):device = torch.device("cuda" if torch.cuda.is_available() else "cpu")input_features = processor(audio_array, sampling_rate=SAMPLE_RATE, do_normalize=True, return_tensors="pt").input_features.to(device)lang_tok = LANGUAGE_TOKENS_WHISPER[LANGUAGE_NAMES[language]]transcribe_tok = processor.tokenizer.convert_tokens_to_ids("<|transcribe|>")notimestamps_tok = processor.tokenizer.convert_tokens_to_ids("<|notimestamps|>")forced_decoder_ids = [(1, lang_tok),(2, transcribe_tok),(3, notimestamps_tok),]predicted_ids = model.to(device).generate(input_features,forced_decoder_ids=forced_decoder_ids,num_beams=1,do_sample=False,)transcription = processor.decode(predicted_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)print(transcription[0])# Get some test audioimport huggingface_hubhuggingface_hub.login()ds = datasets.load_dataset('Sunbird/salt', 'multispeaker-lug', split='test')audio_column = 'audio'ds = ds.cast_column(audio_column, datasets.Audio(sampling_rate=SAMPLE_RATE))audio_array = ds[0][audio_column]['array']# Specify a languagelang = 'Luganda'transcribe_by_whisper(audio_array, lang)# Ekikoola kya kasooli kya kyenvu wabula langi yakyo etera okuba eya kitaka wansi.
The Word Error Rate (WER) and Character Error Rate (CER) are measured for each language on the evaluation datasets. For ach, eng, lgg, lug, nyn, teo, the dev split of multispeaker-* subsets in Sunbird/salt dataset were used for evaluation. For myx and xog, the last 100 examples of train split of makerere-yogera-* subsets in Sunbird/external-speech-data dataset were used. For the rest languages, we picked first 50 examples in the dev split of each subset in huwenjie333/speech dataset.
markdown
code language WER CER n_examplesafr Afrikaans 0.084 0.028 170aka Akan 0.323 0.129 50amh Amharic 0.401 0.265 250bam Bambara 0.361 0.170 149bem Bemba 0.408 0.091 100ber Berber 0.147 0.041 50cgg Rukiga 0.242 0.057 49dag Dagbani 0.321 0.109 100dga Dagaare 0.294 0.113 50ewe Ewe 0.291 0.097 78fra French 0.032 0.019 50ful Fulani 0.395 0.115 149hau Hausa 0.168 0.047 287ibo Igbo 0.340 0.117 247kab Kabyle 0.447 0.219 50kau Kanuri 0.389 0.101 50kik Kikuyu 0.209 0.099 50kin Kinyarwanda 0.307 0.084 100kln Kalenjin 0.471 0.112 100koo Rukonjo 0.527 0.093 47kpo Ikposo 0.617 0.231 50led Lendu 0.273 0.093 50lin Lingala 0.243 0.104 193lth Thur 0.148 0.059 41luo Luo 0.226 0.051 149luy Luhya 0.203 0.038 50mlg Malagasy 0.143 0.060 49nbl Ndebele 0.263 0.051 10nya Chichewa 0.273 0.058 98orm Oromo 0.241 0.064 119pcm Nigerian Pidgin 0.167 0.065 100ruc Ruruuli 0.480 0.105 38rwm Kwamba 0.584 0.234 49sna Shona 0.196 0.050 147som Somali 0.381 0.127 149sot Sotho 0.229 0.117 94swa Swahili 0.087 0.022 99tsn Tswana 0.067 0.018 98ttj Rutooro 0.222 0.063 49wol Wolof 0.319 0.122 135xho Xhosa 0.262 0.054 126yor Yoruba 0.384 0.136 249zul Zulu 0.207 0.042 143ach Acholi 0.197 0.044 101eng English 0.041 0.016 101lgg Lugbara 0.216 0.052 101lug Luganda 0.109 0.022 103nyn Runyankole 0.262 0.051 103teo Ateso 0.275 0.068 101myx Lumasaba 0.347 0.071 100xog Lusoga 0.301 0.056 100
