Toronto-based voice-to-text startup Superwhisper has partnered with Canadian artificial intelligence company Cohere to integrate the Cohere Transcribe speech-recognition model into its dictation platform. The collaboration gives Superwhisper users access to an open-source model that can run directly on personal devices rather than relying on remote cloud processing. Both companies are positioning the integration as an all-Canadian technology stack combining transcription performance, privacy, and local control.
Cohere Transcribe Joins Superwhisper
Cohere Transcribe is now available through Superwhisper’s models library for users across its supported desktop and mobile platforms. Users can download the approximately 1.3-gigabyte model, select it as their voice model, and continue dictating even when an internet connection is unavailable. Because the transcription process takes place locally, users’ recorded speech does not need to leave their device or be transmitted to an external server.
Addressing Speed and Accuracy
Superwhisper said the partnership addresses one of the main challenges facing dictation software: producing sufficiently accurate text without slowing down the user’s workflow. Cohere Transcribe is a two-billion-parameter model supporting 14 languages, including English, French, Spanish, Mandarin, Japanese, and Arabic. The model is designed to deliver strong transcription throughput while remaining compact enough to operate on consumer hardware.
Superwhisper has also connected Cohere Transcribe with its existing vocabulary system, which helps the application recognize words that may not be handled reliably by a standard transcription model. These can include personal names, company names, product terminology, technical jargon, saved snippets, and frequently used expressions. The integration is intended to reduce the time users spend correcting transcripts after speaking.
Competitive Speech Recognition Performance
According to Superwhisper, Cohere Transcribe has achieved competitive results on public speech-recognition benchmarks, including tests involving clean audio, multiple speakers, and far-field recordings. The model is released under the Apache 2.0 open-source license, allowing developers and organizations to inspect, modify, and deploy it under the license’s conditions. Superwhisper currently keeps Nvidia’s Parakeet as its default local download because it covers more languages, although Cohere Transcribe may eventually become the platform’s baseline local model.
Building a Canadian AI Stack
The two companies have emphasized that both Superwhisper and Cohere were built in Toronto, creating a domestically developed technology stack from voice input through transcription. Superwhisper said local processing can give organizations greater control over sensitive audio data while reducing their dependence on foreign-owned cloud infrastructure. The partnership also reflects broader efforts to increase the adoption of Canadian-developed artificial intelligence tools among businesses and institutions.
Expanding Into Enterprise Workflows
Superwhisper initially developed its business around individual users, but enterprise customers have become a growing part of its revenue base. The company told BetaKit that approximately 40 percent of its revenue now comes from enterprise plans, compared with an earlier period when individual users generated between 80 and 90 percent. Working with Cohere could support that transition by bringing Superwhisper’s dictation technology into more corporate environments.
The companies are also exploring deeper connections between Superwhisper and Cohere’s workplace deployments. Such integrations could allow dictated emails or documents to incorporate relevant information from a user’s authorized business data instead of processing each voice command without context. Superwhisper is additionally planning to expand its meeting mode, creating a locally operated alternative to cloud-based meeting transcription and assistant platforms.
Voice as a Workplace Interface
Superwhisper enables users to create emails, messages, documents, and instructions for AI agents through spoken input. Interest in voice-based computing has increased as developers and other professionals use speech to interact with coding assistants, workplace software, and generative AI systems. More accurate local transcription could make voice a more practical interface for completing everyday knowledge-work tasks without exposing audio to external platforms.
The partnership combines Cohere’s speech-recognition technology with Superwhisper’s growing dictation platform as voice becomes a more prominent interface for workplace applications and AI tools. Its immediate value centers on multilingual transcription, offline operation, customized vocabulary support, and greater control over where audio is processed. For both companies, the agreement also provides an opportunity to promote Canadian-built AI infrastructure while expanding their reach among enterprise customers.