kenarsa

Sweet! Go for it!

Agreed. We try to put the price for easy cases. I rather to not put estimates. It probably ends up making customers unhappy.

This is really good feedback. Agreed. What I have seen work is to put common cases. It probably makes our life on Picovoice side easier as answering repeating price questions is not a fun day to spend our time. We will address this soon. I promise.

Numbers and complex words (I am assuming you mean something like "ok blah"?) are doable easily. I am not sure what you mean by unknown words. Could you elaborate?

Obviously, you can just grab the audio stream after the phrase is detected and route it to whatever you like. Google ASR or even a local one running on the device!

Thanks for the feedback :) Hopefully, we find a way to make you like our project!

the project you mentioned is wake-word for ARM only. Great project, BTW. For wake word, we provide on-demand model generation. We do also more than wake word. Finally, we can run on other CPUs/OSs as well.

you are welcome!

Sound classification is in our roadmap in 2019.

Agreed. The reason for not supporting grammar in this product is that we want to keep it extremely lightweight. We do have upcoming products that support grammar as well.

If you have a commercial application we can build you the model for any wake word. Since this requires some engineering work on our side we can only offer it to commercial customers at this point.

The demo is using WebAssembly which is supported by Chrome. It also uses Web Audio API which I believe is again supported by Chrome. I just used the demo on my Android phone using Chrome. I wonder what could be a problem. It the mic on? :) If yes, maybe provide me the version of chrome you are using. I will look into it.

For speech-to-text right? I provided the quote for voice control engine: https://picovoice.ai/#wake-word-detection

The voice control comes in two variations standard and tiny. The tiny one consumes even fewer resources. I provided metrics for the standard one. You can check the benchmark repo on benchmark it yourself as well :) https://github.com/Picovoice/wakeword-benchmark

RPi3 is definitely faster. We also run on RPi 1/zero/etc.

Stay tuned. We will add two more repositories (1) speech-to-intent (2) speech-to-text

BTW, Picovocie runs on RPi (all variant not only RPi3).

We do support Windows!

You can use this for free under Apache 2 license for a personal non-commercial use. No licensing needed. You don't need to pay us a penny as long as you don't make money off it.

I would love to see Picovocie on ESP32. We've looked into this as we get many requests for Picovoice on ESP32. The challenge is to find a commercial request at a reasonable scale to cover the porting effort. I suspect it should be less than a month of work on our side.

You should check this out: https://github.com/Picovoice/wakeword-benchmark

We do work with a couple of SoC manufacturers and will disclose some of the results when our partners are ready. In general, we can run on any MCU with a C compiler and 200KB of RAM (maybe less if there is fast FLASH available). We already of models working on ARM Cortex-M and Cadence's HiFi4.

The reason the quote is not present is that there are just too many factors involved. What is the platform you are using (iOS, Android, ARM Cortex-M, DSP, etc)? What is the scale? Are you deploying to ten devices or ten million? Do you have specific runtime requirements? Some people have limited RAM (e.g. I only have 32KB of RAM), some limited CPU, some limited FLASH, etc. How many models do you need? Are there proper words or brand/made up names? Is it English? How much engineering support do you need?

In reality, it is a lengthy decision tree. I do not want to put that here or on the company's website as it will just maximize the confusion. But it makes sense to put the common easy cases and then ask for contact in the rest. Which is probably the path we take as it will save us some time.

That being said I challenge you to find a company who offers similar tech and have pricing on their website. I suspect what I mentioned is the reason. I could be wrong.

We will open source Cheetah this year with a similar licensing model as Porcupine.

Hello. Thanks for the comment. We have plans to open source two more products this year. with similar licensing compared to Porcupine.

1- Speech-to-Intent: It allows you to issue complex voice commands in a specific domain and in turn returns the intent. For example, in the case of a coffee maker you can say "Please may I have a single shot espresso with no milk and two sugars". The engine returns a JSON-like object with {"product": "espresso", "milk": "no", "sugar", "two", "# shots": "2"}. It is a tightly coupled domain-specific speech recognition and NLU. It is small (less than 3MB and 8% CPU usage on RPi3) and ideal for home automation, industrial application, service industry, etc.

2- Speech-to-Text: It is large vocabulary speech recognition software that runs locally. It will support all platforms currently being supported. It allows you to do large vocabulary transcription with high accuracy locally on an embedded platform.

I am the founder of Picovoice. I am NOT going to comment on comparison as obviously I will be biased ;)

With Picovoice you can use the voice control engine to accomplish this. Maybe something similar to this demo?

https://picovoice.ai/#voice-control-demo

The cool thing about this engine is that it is tiny. It uses less than 8% CPU on RPi3 and altogether it is less than 2MB (code, model, etc). Technically you can run it on something much smaller and cheaper than RPi.

Alternatively, the speech-to-intent engine could be a good candidate. More information on this along with an interactive demo will be released this weekend.

It is true! Have you tried it? :)

We will add another demo for speech-to-intent module fairly soon (over this weekend). Stay tuned!

This is Alireza. I am the founder of Picovoice. Thanks a lot for the information and also sharing. I 100% agree we do need to work on the website and its content. It is definitely in our TODO list. Will prioritize. In the meantime happy to answer any commercial questions via email contact@picovoice.ai

Thanks again for sharing

That's odd. If there are errors in the console. Please open an issue in the GitHub repo is possible: https://github.com/Picovoice/Porcupine

Otherwise, I am thinking it could be something related to the microphone?

Sounds like a great idea. Let us know if you face any problems integrating it. Happy to help. Cheers!

thanks a lot for the feedback :)

Take a look at this standalone demo if you like https://github.com/Picovoice/Porcupine/tree/master/demo/js

it should be a matter of tweaking a few parameters.

Sounds like a good idea. I will think about it. thanks for feedback.

Thanks. Do you use the default parameters and tune from there?