Recent press headlines have highlighted the fact that text messaging is increasingly dominating personal communications. That includes both person-to-person contacts via e-mail and Instant Messaging, as well as business process applications that deliver information or important, time-critical notifications to people.
In fact, two recent federal government initiatives will only reinforce the trend towards text messaging and the need for the flexibility of UC and UM.
- The FCC (News - Alert) approved a plan for a voluntary mobile phone emergency alert system for use by wireless carriers to supplement legacy public broadcasting services. This will enable more delivery of such public warnings to personal mobile devices. Needless to say, more personalized notifications by business applications will also rely on text messaging, which can always be listened to through Text-to-Speech technology by the recipient, whenever necessary.
The key points of such a plan are:
- Emergency notifications will originate from automated detection functions
- Must target individual end users without unnecessarily disturbing non-affected users
- Must be deliverable in a timely manner regardless of location
- Must enable response by the notification recipient in a timely manner regardless of location and device modality
- A new bill was introduced in the House of Representatives banning the use of mobile devices for voice calls on U.S. airplanes while in flight. This bill, called the “HANG UP” act will spare fellow passengers from the “noisome disruption” of mobile conversations that cash-strapped airlines will be tempted to sell. However, text messaging and email access will be allowed.
So, retrieving voice mail messages in text form will fit in nicely with these other text messaging activities. The question, however, is how accurate do the transcriptions have to be and much will such a service cost?
Human Transcription of Voice Mails?
As I pointed out in my last article on the subject, there are many reasons why automatic speech recognition (ASR) will never be perfect for voice messaging, especially from mobile callers, including:
- Unusual words
- Strange dialects and accents
- Noisy environments
- Poor connections with mobile devices
However, consumer voice mail doesn’t really justify the expense of manual transcriptionists, according to long-time speech recognition expert, Walt Tetschner. Walt commented on the use of manual transcriptionists in his March issue of ASRNews as follows:
“The approach that is being used by companies such as SpinVox (News - Alert), SimulScribe and Nuance are flawed and destined for failure. Voicemail-to-text is a very nice convenience feature that will find some solid utilization, but only at a price that is significantly lower than what can be achieved with VM transcriptions that involve human transcriptionists.”
CallWave’s (News - Alert) “Gisting” Technology approach makes sense: The transcription of voicemail with commercially available speech recognizers yield word error rates in the 30% to 40% range (for typical VM messages). This means that any reliable word-by-word transcription service for voicemail would require human transcriptionists to scan, verify and correct the automated output. A large human effort would introduce a large cost, which would grow linearly with adoption of the service, and therefore would almost certainly not be scalable for a volume rollout with a telecom carrier”.
“The CallWave approach appears to be using speech recognition properly. It analyzes the voice mail message and extracts the “gist” of the message. This is precisely what users want. They want immediate notification that they received a voicemail, who it was from, and information that would permit the user to prioritize the message and respond appropriately. It does the VM-to-Text conversion without the use of human transcriptionists. This not only eliminates the cost of the human transcriptionists, but also delivers what the users want.”
Incidentally, the option for the recipient to selectively listen directly to any important message segment that could not be translated will always be useful for the exception cases.
What Do You Think?
Internet Telephony Magazine
Click here to read latest issue
CUSTOMER 
Cloud Computing Magazine
Click here to read latest issue
IoT EVOLUTION MAGAZINE
