Wednesday, March 18, 2009

SpinVox response for Google voice - where is the Nuance response?

As per the recent google voice voicemail transcription announcement, SpinVox is reponding -
In a recent post by Rich Tehrani from TMC, he quote a response from SpinVox about Google voice:

Google is entering a marketplace that continues to be led by SpinVox, the world's largest privately-held speech technology company. We're excited by the launch of Google Voice because it will demonstrate the benefits of speech-to-text conversion and validate its deployment as a network service to an increased audience. We have already launched carrier-grade services with 13 operators - including recently with Skype - on five continents and SpinVox is in use by in excess of 30 million users. SpinVox's 97 percent accuracy in conversion is now the benchmark around the world. - Christina Domecq, co-founder and CEO of SpinVox

And this 97% are human results. I am not sure that Google will post any response but we need to wait for usability feedbacks from users.

I further wonder about Nuance's response to Google Voice..?

7 comments:

Anonymous said...

The quote`And this 97% are human results` may mislead readers into believing SpinVox operates a human transcription service to achieve its benchmark conversion quality levels. This is not the case.

The SpinVox Voice Message Conversion System (VMCS) works by combining state-of-the-art speech technologies with human intelligence and learning.

The majority of the VMCS (voice-to-screen) conversion is carried out by automated speech technologies. VMCS is continually evolving – it is supported by a live-learning process, enabling the system to take on an ever-increasing proportion of automated conversion.

One of the key technology breakthroughs inherent in the SpinVox VMCS is that the automated system `knows what it doesn't know` it is able to refer to a human being for assistance when required.

SpinVox employees support the VMCS system in a variety of ways – from an individual following detailed quality process steps to language laboratory experts employing assurance processes and managing linguistic resources.

I hope this helps clarify our approach.

Ofer said...

Jonathan,
Can you elaborate what percentage of the messages in not reviewed at all by humans (both listening and reading)?

Anonymous said...

Google's move is a Good Thing for the industry. By whacking down to zero the value artificially associated with automatic speech recognition, Google is merely staying one step ahead of the trend -- that plain vanilla ASR is about to become a technical problem sufficiently easy to solve by many players. Google's action is a reminder of the obvious -- that we have to create value by building delightful solutions and services for customers. Resting on technology advantage alone will, at best, buy you a little bit of time.

Anonymous said...

97% good result may be with English, but with other languages as for instance Dutch I have a feeling it will be very difficult to get.

We have been working on Dutch speech to text and only with proper training we can get these kind of figures.

Kind reagrds,
Bert Mengerink
DUTCHEAR

Anonymous said...

I think the reason you don't see a Nuance response is that google hired half the engineers away from Nuance to develop their speech recognition technology. Details are still a little hazy if Nuance technology is being licensed to google or if google did this all on their own from scratch.
My guess is about 80% of voicemails go through a human trnscriber. They had call centers in Pakistan, Kenya, Egypt, South Africa etc. with hundreds of transcribers working for them. Labor is cheap.

Anonymous said...

Ofer - we don’t disclose the exact proportion of the conversion overall that is achieved by machine alone, as that is commercially sensitive information. VMCS can, however, deliver 100 per cent machine conversion of individual messages.

Anonymous said...

jonathan, you don't disclose the exact proportion because it's extremely high.

Bookmark and Share