Twilio voice transcribing replacement
c#, sms, twilio, voice-recognition
Solution
Here are the two (the only that I know of) APIs that offer transcription. Both offer machine and human transcription, depending on your needs:
VoiceCloud: You post a link to the audio file (mp3/wav) and they'll send the text to a callback URL when processed. Transcription is $0.10/minute for machine transcription, $0.009/sec for human assisted. I believe the $250.00 monthly minimum only applies to the human transcription.
QuickTate: You submit an audio file or link via XML-RPC, and QuickTate will send the finished transcription via a callback URL. You can also check job status to get a partial transcription. Price depends on prepay and they offer medical quality as well. It's from around $0.01 to $0.025 per word.
I've used both, and my recommendation would be QuickTate if you need quality transcriptions (although VoiceCloud will still beat the quality of Twilio's native transcription). If you have many short files Quicktate will likely be lower cost, as VoiceCloud (when I used it) had a minimum charge of 30 seconds.
Problem
I'm looking to use Twilio to - Receive a call - Transcribe the voice - Text the user back with the transcribed recording I've got this nicely setup but the transcribing on Twilio is of a bizarrely low quality. It could be that it can't understand my Yorkshire accent (it's a Califorian company after all) but there are complaints from others with presumably clearer speaking voices that it is truly awful. I have an idea that I could swap out their own transcription service and use a third party. I reckon I can grab the mp3 file that is generates. So what i need is - A third party transcription service callable via a webservice or similar - The service generates a callback when done - The service consumes mp3 files - The service actually transcribes reasonably. It doesn't need to be perfect as the text will go back to the user for correction but it does need to be fairly good. Does anyone have experience of such a thing? I'm not committed to Twilio - I'm still at proof of concept so a complete alternative would be of interest as well. Does that seem like a reasonable thing. Have I gone mad event trying it? Thanks for all response. Just for completeness - I'm using an MVC 3 architecture with a Razor engine but I don't think that's hugely relevant. Many Thanks