If you want to convert a series of 1's and 0's to text, ask a computer. If you want to convert a voice message into a text message, ask a human.
That's the crux of the argument given by VoiceCloud, which launched into the voice-to-text fray on Tuesday with a speedy, employee-focused transcription service.
Speech-to-text is a huge trend in wireless and Web apps this year that--judging from the numerous services showcased at the CTIA Wireless conference in Las Vegas this week--is heating up as an important frontier in cell phone software. You may have read about Spinvox and SimulScribe, and CallWave, but according to VoiceCloud CEO Gerald Marolda, his company's service gives a faster, more faithful experience than letting software do linguistics.
With VoiceCloud, the voice message is broken into chunks using an in-house, proprietary software, and each audio segment is sent to an available transcriber who types the section and resubmits it. The software takes over from there, recombining the textual fragments and pushing them, quickly, to an e-mail or SMS message--your choice.
"Many companies claim to use voice-to-text software," said Marolda, when asked why he preferred people power. "But there is no technology in the market that exists right now that gives you the accuracy you need." Instead, VoiceCloud's CEO suggests that many competitors outsource editors or have humans double-check software transcriptions. VoiceCloud's free beta service is available from any cell phone browser and is optimized for the iPhone.
Crowdsourcing gets it right
YouMail's CEO, Alex Quilici, agrees that voice-to-text as it stands is essentially broken. But instead of employing Homo sapien ears like VoiceCloud, he opts for community cooperation and socially-derived ratings as the sure fix, believing that bipeds can be leveraged to correct transcription mistakes over time.
A new feedback loop joins the readable voice-mail service of YouMail, a neat bit of webware for managing voice mail like e-mail. A pop-up feedback form on the online account offers the opportunity for the masses to send programmers ratings, red flags, and helpful edits that will be integrated back into the transcribing process. Oft-repeated errors will be the first to undergo scrutiny, and responsive users giving sound feedback can help speed the correction process, Quilici added.
In addition to releasing the voting function for transcription quality, YouMail also has rolled out SmartGreetings, a feature that personally greets callers whose names match entries in a database of 4,000 "common" names. Maybe YouMail will get people power to grow that database, too.