One of the great things about it is its word-level time stamp and confidence data that it returns...here's a few super cuts I've made from the presidential primary debates:
It's not perfect by any means, but the granular results give you a place to start from...here's a super cut of cuss words from a well known episode of The Wire...only 59 such words were heard by Watson even though one scene contains 30+ F-bombs alone:
It took me a while to understand what you did here. I was waiting for some kind of subtitles showing the recognition ability.
But you are saying you performed speech recognition on the full video then edited it according to where the words you targeted were found. I liked the bomb/terrorist one, the others didn't seem to be "saying" anything.
Yeah, I was a bit lazy...I could have used moviepy (which I currently use, but merely as a wrapper around ffmpeg) to add subtitles to show which identified word was identified...I'm hoping to make this into a command-line tool for myself to quickly transcribe things...though making supercuts is just a fun way to demonstrate the concepts.
The important takeaway is that the Watson API parses a stream of spoken audio (other services, such as Microsoft's Oxford, works only on 10-second chunks, i.e. optimized for user commands) and tokenizes it...what you get is a timestamp for when each recognized word appears, as well as a confidence level and alternatives if you so specify. Other speech-transcription options don't always provide this...I don't think PocketSphinx does, for example. Or sending your audio to a mTurk based transcription service.
Here's a little more detail about The Wire transcription, along with the JSON that Watson returns, and a simplified CSV version of it:
It limits input audio to 2 minutes. And I would have to guess that its model is specifically tuned to phone messages, i.e. one speaker, relatively clear and focused audio, and certain probabilities of phrases.
Kids, it's called "speech recognition". Voice recognition also exists, but it's the task of identifying a user based on his/her voice, not the task of transcribing spoken input as text.
Are there any decent opensource projects out there (preferably with Python APIs) that do speaker or "voice recognition" reasonably well? I know this is an area of active research in academia.
It really would be amazing to be able to get voice recognition software that covers at least recognizing a small enough fraction of our language to be useful without having to reach the cloud. It is definitely a dream I hope we one day achieve, thanks for the article, will test it on my day off and play with it a bit.
Pocketsphinx/Sphinx with a small, use-case specific dictionary showed much better accuracy for my accent and speech defects, than any of these cloud based recognition systems. I used a standard acoustic model, but it probably would have been even more accurate had I trained a custom acoustic model.
For simple use cases like home automation or desktop automation, I think it's a more practical approach than depending on a cloud API.
I haven't tried out Pocket Sphinx myself...could you describe the training process, e.g. how long did it take, how much audio did you have to record, how easy was it to iterate to improve accuracy?
PocketSphinx/Sphinx use three models - an acoustic model, a language model and a phonetic dictionary.
I'm no expert, but as I understand them, the acoustic model converts audio samples into phonemes(?),
the language model contains probabilities of sequences of words, and the phonetic dictionary is a mapping of words to phonemes.
Initially, I just used standard en-us acoustic model, US english generic language model, and its associated phonetic dictionary.
This was the baseline for judging accuracy. It was ok, but neither fast nor very accurate (likely due to my accent and speech defects).
I'd say it was about 70% accurate.
Simply reducing the size of the vocabulary boosts accuracy because there is that much less chance of a mistake. It also improves recognition speed.
For each of my use cases (home and desktop automation), I created a plain text file with the relevant command words.
Then used their online tool [1] to generate a language model and phonetic dictionary from it.
For the acoustic model, there are two approaches - "adapting" and "training".
Training is from scratch, while adapting adapts a standard acoustic model to better match personal accent or dialect or speech defects.
I found training as described [2] rather intimidating, and never tried it out. This is likely to take a lot of time (a couple of days atleast I think, based on my adaptation experience).
Instead I "adapted" the en-us acoustic model [3].
About an hour to come up with some grammatically correct text that included all the command words and phrases I wanted.
Then reading it aloud while recording using Audacity. I attempted this multiple times, fiddling around with microphone volume and gain,
trying to block ambient noise (I live in a rather noisy env), redoing it, final take. Took around 8 hours altogether with breaks.
Finally generating the adapted acoustic model. About an hour.
About 95% of the time it understands what I say. About 5% of the time, I have to repeat. Especially with phrases.
Did this on both a desktop and raspberry pi. The Pi is the one managing home automation. I'm happy with it :)
If not confidential, can you describe what kinds of automation you used this for, particularly the desktop automation?
I was interested in automating transcription to text of my own reminders to myself and other such audio files, say taken on the PC or on a portable voice recorder, hence the earlier trials I did. But at the time nothing worked out well enough, IIRC.
Nothing confidential at all :). I was playing with them because I personally don't like using keyboard and mouse, and also have some ideas for making computing easier for handicapped people.
My current desktop automation is doing command recognition. Commands like "open editor / email / browser", "shutdown", "suspend"...about 20 commands in all. 'pocketsphinx_continuous' is started as a daemon at startup and keeps listening in the background (I'm on Ubuntu).
I think from a speech recognition internals point of view transcription is more complex than recognizing these short command phrases. The training or adaptation corpus would have to be much larger than what I used.
He he, the voice "shutdown" command you mention reminds me of a small assembly language routine that I used to use to reboot MSDOS PCs; it was just a single instruction to jump to the start of the BIOS (cold?) boot entry point, IIRC (JMP F000:FFF0 or something like that). Used to enter it into DOS's DEBUG.COM utility with the A command (for Assemble) and then write it out to disk as a tiny .COM file. (IOW, you did not even need an assembler to create it.)
Then you could reboot the PC just by typing:
REBOOT
at the DOS prompt.
Did all kinds of tricks of the trade (not just like that, many other kinds), in the earlier DOS and (more in) UNIX days ... Good fun, and useful to customers, many a time, too, including saving their bacon (aka data) multiple times (with, of course, no backups by them).
My impression is that the super accurate stuff like Google's voice recognition and Siri are all feuled by masssssssive amounts of data. So you build up these recognition networks based off of a bunch of data sources, and get better over time, but the recognition is more based off of the data than the code.
It's the whole "Memory is a process, not a hard drive" thing: Voice recognition as it is today is a slowly evolving graph from input data. You could in theory compress the graph and have it available offline. But it would be hard to chop it up in a way that doesn't completely bust the recognition.
There's actually some research on compressing ANN's to the size that it could be embedded in all sorts of devices. I think I saw something about it on HN a few months back?
> It really would be amazing to be able to get voice recognition software that covers at least recognizing a small enough fraction of our language to be useful without having to reach the cloud.
Well, I guess at some point this functionality will become part of the OS. When OSX and Windows offer this, then Linux cannot stay behind, and we will see open source speech recognition libraries.
There are plenty of those. Voice recognition is nothing new—I remember playing around with "speakable items" back in Mac OS 7. It did well enough to recognize certain key words and phrases.
Every Mac with OS 10.9 and later comes with speech recognition software that works without internet access and is really really good. You can dictate entire documents, emails, even have it type commands in terminal etc. In OS X 10.11 you can even drive the UI by only speaking.
> It really would be amazing to be able to get voice recognition software that covers at least recognizing a small enough fraction of our language to be useful without having to reach the cloud.
Are there any academic groups working on this topic, and do they have prototype implementations?
Julius [1] is a pretty good offline speech recognition engine. In my tests it seems to have about 95% accuracy in grammar-based models, and it supports continuous dictation. There is also a decent Python module which supports Python 2, and Python 3 with a few tweaks.
HOWEVER:
The only continuous dictation models available for Julius are Japanese, as it is a Japanese project. This is mainly an issue of training data. The VoxForge models are working towards releasing one for English once they get 140 hours of training data (last time I checked they were around 130); but even so the quality is likely to be far less than commercial speech recognition products, which generally have thousands of hours of training.
Julius is my preferred speech recognition engine. I've built an application[0] which enables users to control their Linux desktops with their voices, and uses Julius to do the heavy lifting.
After a quick look, it seems Julius doesn't use the new deep-learning stuff?
In terms of data,
http://www.openslr.org/12/
says it has 300 hours + of speech+text from librivox audiobooks. Using Librovox recordings seemed a great idea for making a freely available large dataset.
Don't expect this to be anything like modern "good" speech recognition. Sphinx is definitely from the 00's when it seemed like speech recognition would never be solved.
Apparently Kaldi is a lot better, but good luck setting it up!
Another project along similar lines is the Jasper Project[0], which has received some HN coverage in the past several years[1]. It interfaces with many of the same speech recognition and text-to-speech libraries.
Very cool! I just started playing with speech recognition in Python for home automation this week. I'm controlling some WeMo switches and my PC with an Android Tablet using Autovoice, and it works well as a proof-of-concept, but Autovoice doesn't always register commands, and the "Okay, Google" speech to text can be slow sometimes. I'd like it to take less than 5 seconds between saying "TV Off" and the TV actually turning off., with Autovoice it's anywhere from 3s to 25s depending on the lag. I also figure with real code, I can get commands that are more flexible than Autovoice's regex.
Aside from circumventing lag, I can also give it some personality. I want to name it Marvin, after the robot from H2G2, so that I can say:
"Marvin, turn the TV off"
"Here I am, brain the size of a planet, and you ask me to turn off the tv. Call that job satisfaction, 'cause I don't."
Does Kaldi need Windows? I only saw installation instructions for Windows. Also .. I just tried Pocket Sphinx ... says it works on Windows and Linux. So .. no non-apple or cross platform speech rec for us mac devs?
AFAIK, On the contrary. They officially support only linux. But community provides Windows support. I do not know Mac support. AFAIK again, Kaldi is mainly targeted server and desktop applications
Nice work, ggulati. I had done some roughly similar stuff, but more basic, using same / similar libraries (but you have researched more libs), a while ago:
Recognizing speech (speech-to-text) with the Python speech module
Excellent post. Very interesting. I see how it works but am using Python 2.7 so based on your headline I suppose it won't work for me. This is the first real lead I've seen for integrating it easily. Pricing isn't terrible, if it goes production. Too bad there is no way to test it first for development. But we're lucky to have this at all.
Most of the stuff I found was for Python 2.7! I'll edit that into the post. My focus was for finding libraries that worked with new Python code, e.g. Python 3.5 code.
All of those libraries have Python 2.7 versions. Actually for all of them you pip install the same library; for pyttsx, `pip install pyttsx` and ignore jpercent's update.
I'm not sure what you mean about pricing and testing for development. Are you referring to Google's services? They offer 50 reqs/day for voice recognition on a free developer API key (https://www.chromium.org/developers/how-tos/api-keys). Google Translate can also be used by gTTS; it will rate limit or block you if you send too many reqs/min or per day without an appropriately registered API key, but you could play around with it for sure.
If voice recognition is important, it might be worth investigating Sphinx more and putting the time to tweak their English language model files. Synthesis is more difficult, though I think the Windows SAPI, OSX NSSS, and ESpeak on *nix are all "good enough." There are also a range of commercial libraries.
I too thought it was Python 3 only before I read it. Maybe a better title would be "Coding Jarvis in Python in 2016" and then explaining in the first paragraph that this is Python 2 and 3 compatible, with your personal focus on 3?
Does this work without an internet connection (once downloaded)? If yes, How big is the downloaded footprint? I still haven't gone through the webpage carefully.
If you use Sphinx for speech recognition and use pyttsx for text to speech (Windows Speech API, OSX NSSS, or ESpeak on Linux) it all works offline - see the "Jarvis's Brain" section.
https://github.com/dannguyen/watson-word-watcher
One of the great things about it is its word-level time stamp and confidence data that it returns...here's a few super cuts I've made from the presidential primary debates:
https://www.youtube.com/watch?v=VbXUUSFat9w&list=PLLrlUAN-Lo...
It's not perfect by any means, but the granular results give you a place to start from...here's a super cut of cuss words from a well known episode of The Wire...only 59 such words were heard by Watson even though one scene contains 30+ F-bombs alone:
https://www.youtube.com/watch?v=muP5aH1aWUw&feature=youtu.be
The service is free for the first 1000 minutes each month.