Speechbot
| Company type | Research project |
|---|---|
| Industry | Internet search engine, Streaming media |
| Founded | 1999 |
| Founder | Compaq research labs |
| Defunct | 2005 |
| Fate | Shut down following closure of HP Cambridge research lab |
| Headquarters | Cambridge, Massachusetts , United States; Australia |
| Products | Streaming media search engine |
| Parent | Compaq (later HP) |
SpeechBot functioned as a web search engine specifically designed for streaming media content and was created within the research facilities of Compaq, which later became part of HP, located in both Cambridge, MA and Australia. The platform was unveiled by Compaq at the Streaming Media West conference held in 1999 in San Jose, CA. The indexed internet radio programs featured titles such as The Motley Fool, Fresh Air, Talk of the Nation, The Dr. Laura Program, and Dreamland hosted by Art Bell. By the middle of 2003, SpeechBot’s index encompassed more than 17,000 hours of multimedia material. The service ceased operations in 2005 following the shutdown of HP’s Cambridge research lab.
The indexing process for SpeechBot involved utilizing a network of Windows workstations to capture streaming audio, while a Linux cluster equipped with speech recognition technology transcribed the sound files. The infrastructure supporting the web server, search indexing, and metadata repository operated on AlphaServer systems running Tru64 UNIX.
When available, existing transcripts were synchronized with the audio streams; if transcripts were absent, approximate text was produced using the Calista speech recognizer, which was based on Sphinx-3. Although the word error rates were elevated due to the poor quality of streaming audio, the search functionality still effectively retrieved pertinent content. Users could access search results that linked directly to the exact position within the stream, allowing them to navigate straight to relevant segments without needing to listen through entire broadcasts.