DOHA: The UK-based company, Speechmatics, will use technology developed by Qatar Computing Research Institute (QCRI), of Hamad Bin Khalifa University, to take Arabic speech-to-text services to its global customers.
The Cambridge-based Speechmatics will use QCRI product — QCRI Advanced Transcription System (QATS) — to transcribe Arabic broadcasts and audio files into text and subtitles. QATS can transcribe modern standard Arabic and four major Arabic dialects — Egyptian, Levantine, North African and Gulf Arabic.
Dr Ahmed Elmagarmid, Executive Director, QCRI, said Qatar is leading global research in speech technology for Arabic. “This is not just a technology transfer — it will allow information-sharing in Arabic around the world,” he said.
Dr Tony Robinson, Chief Scientific Officer, Speechmatics, said the development, which used the company’s Auto-Auto framework, would ensure Arabic-based content is “more discoverable and easily consumed.
“Speechmatics will help QCRI expand its reach to a broad range of industries and geographies with market-leading speech-to-text services based on latest research in machine learning and artificial intelligence,” he said.
QCRI Arabic Language Technologies Principal Engineer Ahmed Ali, who has been leading the speech team, said Deep Neural Network (DNN) and Recurrent Neural Network (RNN) architecture were used in QATS. “We used over 2,000 hours of Arabic speech to develop and train QATS, in addition to a large archive of the Web 2.0 content.”
He said the agreement would give QCRI access to diverse data which would enable it to hone its Arabic language technologies research.
An independent media monitoring company in May 2015 found QATS consistently outperformed leading competitors in standard and dialectal Arabic benchmark tests by at least 10 percent. It also won the ‘Best in Show’ award at the third edition of the BBC’s #NewsHACK in December 2014 for translating BBC Arabic videos into English, including subtitles, and voiceover using speech synthesis.
Al Jazeera Media Network has been using versions of QATS to transcribe its daily Arabic news reports for almost two years. Over 3,000 hours have been transcribed using the product so far. Arabic is the world’s fourth most popular language spoken in at least 60 countries.
The Peninsula
