FOOD FOR THOUGHT IBM Watson STT - E-mail Template

[John C. Reid]

I have the transcriptions working, and they work well save two minor issues.

Each line of the transcription has a double quote and a comma at the end of the line, ie ", and I suspect this might me a typo somewhere in the script that is either missing a quote or it has an early close quote, but I can't find it.

The script automatically appends the transcription to the end of the E-mail template. I use this to generate support tickets and I would prefer to have the transcription come first, or better yet have it turned into a variable that I can put anywhere I wish in the e-mail template.

Any assistance is appreciated.

 

[John C. Reid]

I have done a little more digging on the IBM script and understand the gist of it. Enough that I am seriously considering moving forward with creating an AWS Transcription version. If I get any spare time I will work on it and report back with my results.

 

[jerrm]

Google around, I think some are already out there.

It's been years, but my testing with AWS Transcription wasn't very good. Biggest issue was speed. The other offerings were 2x or more faster. I didn't keep the benchmarks, but I think it was a lot more than 2x, IIRC. Surprised me considering the experience they should have with Alexa.

The other issue for me was the API - no simple curl API, the SDK was pretty much required. Not a big deal if already hitched to the AWS platform, but I'm not. I don't add an extra dependencies if I can avoid it, and Google/IBM/MS have easy curl APIs. Didn't see any need to expand the script further to add AWS as a fourth option (actually fifth at this point - my script does deepspeech too).

 

[John C. Reid]

I am already familiar with the AWS CLI tools, and the only dependency for AWS CLI is Python, which is already there.

Hopefully speed won't be an issue. The issues I am having with Watson is the total lack of natural language. In addition to getting words wrong (in that aspect it seems no better or worse than Nexiwave has been) it seems to only work on translating a word to a word with no concept of grammar. Numbers are spelled out rather than being digits, there is zero punctuation, it is almost unusable to me.

Not having used AWS Transcription yet I don't know what I will get, but they purport to understand natural language. The transcription should come pack with proper sentence structure including capitalization and punctuation. Number should be normalized to digits. It can use context to determine words if they are ambiguous, etc. Also, I can control some aspects of the transcription from the AWS dashboard and even add some of my own words and phrases if it tends to get them wrong frequently.

 

[jerrm]

the only dependency for AWS CLI is Python

AWS CLI itself is a dependency I don't otherwise need, but I'd have gladly added the dependency if there had been any meaningful improvement over the IBM/MS/Google offerings.

The watson formatting issues sound off. After checking my IPBX test system, it looks like Ward is not enabling smart_formatting. Try adding smart_formatting=true to the URL query string.

Even with smart_formatting, formatting is one of the weaker IBM areas. MS and Google are generally much better in that regard, wouldn't be surprised if AWS is too.

My personal VM runs through IBM/MS/Google and Deepspeech. The three online services are all comparable, I keep deepspeech in the mix as a curiosity. I'd give the overall ranking as 1:Google/2:MS/3:IBM, but that is largely subjective.

Overall word accuracy seems very comparable, but different. It's not unusual to see a transcription where two services get a phrase completely wrong and the third is spot on, only to have another phrase in the same message with exactly the opposite results.

 

[dhoppy]

Forgive the hijack, but do your clients worry about privacy and sending VM to a 3rd party for transcription?

 

[jerrm]

Forgive the hijack, but do your clients worry about privacy and sending VM to a 3rd party for transcription?

Not so far. It's made clear to the client and is signed off on. The traffic is all encrypted. I use the most secure settings available. There shouldn't be anything retained (if you trust the service's claim).

 

[DoctorJ]

I have the transcriptions working, and they work well save two minor issues.

Each line of the transcription has a double quote and a comma at the end of the line, ie ", and I suspect this might me a typo somewhere in the script that is either missing a quote or it has an early close quote, but I can't find it.

The script automatically appends the transcription to the end of the E-mail template. I use this to generate support tickets and I would prefer to have the transcription come first, or better yet have it turned into a variable that I can put anywhere I wish in the e-mail template.

Any assistance is appreciated.

Sorry for the late reply. Any way to remove the ", at the end of every line?

 

[AndyInNYC]

AWS CLI itself is a dependency I don't otherwise need, but I'd have gladly added the dependency if there had been any meaningful improvement over the IBM/MS/Google offerings.

The watson formatting issues sound off. After checking my IPBX test system, it looks like Ward is not enabling smart_formatting. Try adding smart_formatting=true to the URL query string.

Even with smart_formatting, formatting is one of the weaker IBM areas. MS and Google are generally much better in that regard, wouldn't be surprised if AWS is too.

My personal VM runs through IBM/MS/Google and Deepspeech. The three online services are all comparable, I keep deepspeech in the mix as a curiosity. I'd give the overall ranking as 1:Google/2:MS/3:IBM, but that is largely subjective.

Overall word accuracy seems very comparable, but different. It's not unusual to see a transcription where two services get a phrase completely wrong and the third is spot on, only to have another phrase in the same message with exactly the opposite results.

My system uses /usr/local/sbin/sendmailmp3 (sendmailmp3.ibm); can you tell me what/where to edit to get smart formatting to work. I'd like to give it a try.

Andrew

 

[kenn10]

There should be a sendmailmp3.ibm in your /usr/local/sbin directory. You can add credentials to it, move your current sendmailmp3 to a new filename and then copy the sendmailmp3.ibm to sendmailmp3. After you get set up on IBM Watson, you need the API key and the url to use. Use Ward's tutorial on Nerd Vittles tutorial at http://nerdvittles.com/?p=32202 . IBM's website has changed somewhat from that but you can figure it out. IBM no longer requires credit card info if you want the free account which is limited to 500 minutes/month.

 

[AndyInNYC]

Sorry, @kenn10 - I guess I was unclear. I'm using the script with IBM (and have been for some time); I'm asking what I need to modify in the script in order to enable smart formatting; i.e. add "this" line "here" where someone tells me what "this" is and where "here" is.

TIA

Andrew