Cohere releases open-source Arabic speech recognition model
Cohere has released a new open-source model called Cohere Transcribe Arabic, built specifically for Arabic speech recognition. The 2-billion-parameter ASR model is, according to Cohere, the most accurate open-source Arabic speech-to-text system currently available.
The model targets the unique challenges of transcribing Arabic speech. These include the wide variety of dialects across different regions, bilingual conversations where speakers switch between Arabic and English, code-switching within a single sentence, and specialized vocabulary used in fields like medicine, law, or technology.
Performance compared to existing models
Cohere says that Cohere Transcribe Arabic outscores Whisper Large V3, the standard Cohere Transcribe model, and other systems in benchmarks. Human ratings of Arabic transcripts on a scale of 1 to 5 showed that the new model scored higher in overall quality, dialect faithfulness, and code-switching accuracy compared to those alternatives.
Cohere provided an image illustrating these human ratings, showing Cohere Transcribe Arabic outperforming both Whisper Large V3 and the standard Cohere Transcribe model in all three categories.
Availability and licensing
Stay updated
Get the day's AI and automation news in your inbox. No spam, unsubscribe anytime.
The model is released under the Apache 2.0 license, making it free to use, modify, and distribute. It is available for download on Hugging Face and can also be accessed through the Cohere API. More benchmarks, example transcriptions, and technical details are available on the Cohere blog.
Cohere is a Canadian AI company founded in 2019, known for its large language models and enterprise AI solutions. The company has focused on making its models available both as open-source releases and through cloud APIs, offering flexibility for researchers and businesses.
Significance for Arabic language technology
Arabic speech recognition has historically lagged behind English due to the language's complex morphology, numerous dialects, and lack of labeled training data. Open-source models like Cohere Transcribe Arabic could help accelerate development of Arabic-language applications in customer service, healthcare, media transcription, and education.
The release underscores a growing trend among AI companies to release specialized models for underrepresented languages, rather than relying on general-purpose models that may not capture linguistic nuances accurately.
Cohere has not disclosed the specific training data used for this model, but the company says it focused on representative samples of various Arabic dialects and code-switched speech.
Related on Neura Market: