Multilingual Video Transcription System
A multilingual transcription and search system that turns raw video into diarized, translated, searchable transcripts.
Speech-to-Text Development in Toronto
A transcription app gives you a text file. The speech systems our ML lead builds go further: who said what, an English translation, and a way to find the part of the recording you need.
Below are three speech projects: a video analysis prototype for MTV, a multilingual transcription and search system for War of Words, and speech research for TietoEvry.
Speech and audio projects
Video Analysis Prototype
A prototype that transcribes, summarizes, and analyzes large volumes of video content to generate metadata automatically.
Speech & Face Recognition R&D
Research and development work on speech synthesis and recognition, plus deep-learning face recognition backed by a vector database.
These projects were built by Pavlo Holubchykov, Raw Code's Lead ML/AI Developer, some of them before Raw Code was founded. More about Pavlo.
What goes into a speech pipeline
Transcripts that show who's speaking
War of Words' pipeline runs on AzureML. WhisperX turns speech into text, and PyAnnote adds speaker diarization, so the transcript marks where one speaker stops and the next starts.
Multilingual Video Transcription System case studyTranslation in the same pipeline
The same system translates transcripts into English through the ChatGPT API, with prompts written to keep each passage's context. Translation isn't a separate step someone runs afterwards.
Multilingual Video Transcription System case studyFinding the right segment
Audio embeddings sit in a MongoDB vector database, so staff can search for segments similar to the one they have, across languages, instead of scrubbing through hours of video.
Multilingual Video Transcription System case studyFrom transcript to metadata
For MTV, the prototype transcribed video with WhisperX, then used the ChatGPT API to summarize and analyze the text and produce metadata automatically. It ran as a FastAPI backend in Docker on Azure.
Video Analysis Prototype case studyProcess and what's included are on our AI Automation Consulting & Implementation page.
FAQ
What is speaker diarization?
Diarization labels which speaker said each part of a recording. A plain transcript runs everyone together; a diarized one splits it by speaker. The War of Words system uses PyAnnote for this.
Which languages can it transcribe?
Whisper-based models cover dozens of languages, but accuracy depends on the language and the audio quality. The War of Words system handles video in several languages and translates the transcripts into English. We test on a sample of your recordings before quoting.
How much does custom speech-to-text cost?
Speech projects fall under our AI automation work, which runs $3k–$30k project or retainer. A transcription pipeline over one source is at the low end; adding diarization, translation, and search moves it up.
Can the audio stay on our own servers?
Transcription can. WhisperX is open-source and can run on infrastructure you control. Steps like translation through the ChatGPT API send text to a third party. Where each step runs is decided at scoping, before anything is built.
Are recordings of people personal information?
They can be, under Canadian privacy laws such as PIPEDA and Quebec's Law 25. What gets stored, for how long, and who can search it should be checked against your own obligations before the system goes live.
More of our work in Toronto
Have a pile of recordings nobody has time to go through?
Send a short sample and we'll tell you on a free call what a pipeline for it would take.
Book free consultation