I've been playing with SpinVox Create, the new open API for the SpinVox speech-to-text (STT) service, and it is wicked cool. Take any recording (well, any short recording), convert it to a uLaw encoded wave file. post it to the API and within roughly four minutes you have the text. It's not always dead on ('Digium' became 'Dig-em!') but the transcription is really very good.
I've built a rough but fairly feature complete Asterisk-to-SpinVox integration in Python. I hope to release it to the public shortly. My goal is to completely displace voicemail. Instead of having a 'voicemail box' your voicemail will become simply another sub-type of your email. Messages will arrive with the text pre-transcribed but with the audio file available if you need to review. My initial cut isn't ready for use in Google Voice but its working and it takes something like 3 lines of dialplan to implement.
Check back soon for a link to the article, full review and the Python source.
Wednesday, April 22, 2009
Subscribe to:
Post Comments (Atom)
1 comments:
Main issue is the cost. You're looking at 70 cents for a one minute long voicemail message using the spinvox API.
Try VoiceCloud's API, or better yet, just use the email gateway. Have asterisk send the voicemail with audio attachment to the gateway address, it'll be converted to text and emailed back.
Post a Comment