7 ms·

> Basically, I want an Alexa setup without the dependence on (or financial expectations from) Amazon Amen. +1 to timers, schedules, integration with a lightin

> Basically, I want an Alexa setup without the dependence on (or financial expectations from) Amazon Amen. +1 to timers, schedules, integration with a lighting bridge or even kitchen appliances e.g. Home Connect, custom routines, music (including Bluetooth and/or TuneIn), line out (so I can connect my own speakers), microphone relays that I can put all around the house instead of owning 5 or 6 RaspPis and the associated 15 to 30 watts Better than math, for me, would be a Wikipedia summary readout that voices the first paragraph (or a reasonably defined "summary length") from simple.wikipedia.com, although metric/imperial conversions are also very handy. Weather integration would also be delightful---including a way to tie this in to reminders e.g. "bring the plants indoors tonight", "remember rain clothes for riding home this evening", "tomorrow is probably the last good sled day until next winter" Oh. And properly bilingual. We have a Philips light named "Turtle". Alexa absolutely cannot handle this name if the input is set to either German or German/English.

kkielhofner 3y ago
> microphone relays that I can put all around the house instead of owning 5 or 6 RaspPis and the associated 15 to 30 watts I've always found this approach odd and this is one the reasons we primarily target the ESP-BOX-3 from Espressif[0]. For $50 you take it out of the box, flash from our web flasher, and stick it in your house. Given that it's ESP32-S3 based total power consumption is somewhere in the 100 milliwatt range. Once initially flashed all upgrades are done over the air (with < 5MB binaries) from our web management interface. All configuration changes are also centrally managed there. Make change, click "Save and Apply", goes live on any number of associated devices in seconds. The thought of managing half a dozen full Linux machines (Pis) just to get audio for stuff like this boggles my mind. Additionally, the audio quality is significantly better with the ESP-BOX-3 vs Pi, etc because the entire device and software stack is holistically engineered for high quality far-field voice capture. Acoustically tuned enclosure, dual microphones, acoustic echo cancellation, blind source separation, voice activity detection, wake word, etc. As I like to tell people - if you think you're going to slap a random microphone together with a Raspberry Pi and compete with Alexa you have another thing coming. I go into greater detail in this issue[1]. Getting high quality audio is step one. Step two is accurate transcription of that audio. Steps 3 - ? are what you're describing. We know what you said, now what do we do with it? Now that we feel steps one and two are more-or-less nailed steps three and beyond are our next focus. The good news is the final step is of course user feedback and we have that handled with text to speech and output on the BOX-3 LCD display. It's just currently the wide gulf in between these that needs focus. In short, our mouth and ears work well. It's the brain that needs work. [0] - https://www.espressif.com/en/news/ESP32-S3-BOX-3 https://www.espressif.com/en/news/ESP32-S3-BOX-3 [1] - https://github.com/toverainc/willow/issues/317#issuecomment-1774172563 https://github.com/toverainc/willow/issues/317#issuecomment-...

kkielhofner 3y ago

> microphone relays that I can put all around the house instead of owning 5 or 6 RaspPis and the associated 15 to 30 watts I've always found this approach odd and this is one the reasons we primarily target the ESP-BOX-3 from Espressif[0]. For $50 you take it out of the box, flash from our web flasher, and stick it in your house. Given that it's ESP32-S3 based total power consumption is somewhere in the 100 milliwatt range. Once initially flashed all upgrades are done over the air (with < 5MB binaries) from our web management interface. All configuration changes are also centrally managed there. Make change, click "Save and Apply", goes live on any number of associated devices in seconds. The thought of managing half a dozen full Linux machines (Pis) just to get audio for stuff like this boggles my mind. Additionally, the audio quality is significantly better with the ESP-BOX-3 vs Pi, etc because the entire device and software stack is holistically engineered for high quality far-field voice capture. Acoustically tuned enclosure, dual microphones, acoustic echo cancellation, blind source separation, voice activity detection, wake word, etc. As I like to tell people - if you think you're going to slap a random microphone together with a Raspberry Pi and compete with Alexa you have another thing coming. I go into greater detail in this issue[1]. Getting high quality audio is step one. Step two is accurate transcription of that audio. Steps 3 - ? are what you're describing. We know what you said, now what do we do with it? Now that we feel steps one and two are more-or-less nailed steps three and beyond are our next focus. The good news is the final step is of course user feedback and we have that handled with text to speech and output on the BOX-3 LCD display. It's just currently the wide gulf in between these that needs focus. In short, our mouth and ears work well. It's the brain that needs work. [0] - https://www.espressif.com/en/news/ESP32-S3-BOX-3 https://www.espressif.com/en/news/ESP32-S3-BOX-3 [1] - https://github.com/toverainc/willow/issues/317#issuecomment-1774172563 https://github.com/toverainc/willow/issues/317#issuecomment-...