Summary: Voiceitt successfully completed a pilot with deaf and hard-of-hearing participants that demonstrated the effectiveness and inclusivity of its voice AI technology in recognizing non-standard speech patterns.
Takeaways:
- Voiceitt’s technology, which includes a unique database of non-standard speech samples, has shown high accuracy in recognizing and understanding speech from deaf and hard-of-hearing individuals, achieving more than 90% accuracy after participants trained the system with 200 recordings.
- The pilot emphasized Voiceitt’s commitment to inclusivity by tailoring its speech recognition software to accommodate the specific needs of deaf and hard-of-hearing users, enhancing their ability to interact with voice-enabled devices and reducing reliance on third-party interpreters, especially in sensitive or private conversations.
- The successful pilot, facilitated by Newlab, sets the stage for further strategic collaborations that could expand the impact of Voiceitt’s technology, potentially transforming how voice AI is used in the deaf and hard-of-hearing community and beyond.
Voiceitt, an Israel-based provider in speech recognition technology for non-standard speech, announced the successful completion of a joint pilot validating its inclusive voice AI technology with deaf and hard-of-hearing individuals. It demonstrates the flexibility and inclusivity of next-generation speech recognition technology for individuals who are English-speaking deaf and hard-of-hearing with non-standard speech.
About Voiceitt’s technology
Voiceitt’s technology was built with and for people with non-standard speech, and its unique speech database includes speech recordings from people with impaired and atypical speech to augment and improve Voiceitt’s voice AI models. This database of non-standard speech allows Voiceitt to provide a more inclusive and accurate experience. Voiceitt further improves its recognition of non-standard speech by providing personalized recognition for individuals who provide a minimum of 200 recordings.
The pilot was facilitated by Newlab, a deep tech venture platform that works closely with innovators as they build, pilot, and scale frontier technologies. The goal of the pilot was to demonstrate an improved accuracy with Voiceitt’s speech recognition with deaf accented speech and provide qualitative insights to evaluate impactful use cases for voice AI among deaf and hard-of-hearing individuals.
“The Voiceitt pilot is directly in line with the goals of the Studio to pilot with startups working on impact-driven advancements in language services,” says Jamie LePinnet, the director at Newlab who facilitated the pilot. “The success of the pilot sets the stage for strategic collaborations that can have a far-reaching impact on customers, and the community at large.”
While the adoption of voice enabled smart speakers and other devices has increased, people with non-standard speech due to being deaf or hard of hearing are often not able to access and enjoy these devices, which pilot participants noted “rarely understands their speech”. Aside from the convenience of speech recognition over typing, some participants noted that interpreters working with deaf individuals do not always accurately represent what the speaker aims to express. Another pilot participant explained that when individuals are discussing a sensitive topic, they may prefer not to have a third-party interpreter present, and inclusive speech recognition would help prevent “communication breakdown” and help them be better understood.
“Speech recognition technology is incredibly valuable for the deaf community,” says Tyler Pujeda, a heart researcher in Boston who participated in the pilot and is now an active user of Voiceitt’s application. “New inclusive products can help us feel listened to and understood.”
Additional Details of the Pilot
The pilot also analyzed quantitative results of Voiceitt’s proprietary speech recognition software, and compared Voiceitt with the industry’s leading providers of speech recognition, measuring accuracy using the word error rate (WER) metric. deaf users in this pilot obtained an average of 8% WER (more than 90% accuracy) after training the system with 200 recordings, according to the company. Voiceitt’s system also achieved higher accuracy than leading industry software for the deaf speakers with non-standard speech who participated in the pilot, even with no training at all.
Rachel Levy, PhD, a speech pathologist and Voiceitt’s customer success manager, emphasized that while Voiceitt has historically focused on individuals with a diverse range of speech impairments, this pilot offered a unique opportunity.
“This pilot validated Voiceitt’s compatibility and increased accuracy with Deaf voices,” Levy says. “However, its greatest impact resulted from the feedback and insights collected from pilot participants which has directly impacted the development of solutions geared towards the needs and preferences of the Deaf community and the other communities we serve and support.”
Photo: Dreamstime
