12 Jul 2024 |
joergz | When STT server is working all cores are at 99.9 percent but temperatur is up to 70 degrees Celsius. | 12:07:56 |
goldyfruit | 0x80000 is a software temperature limit | 12:14:14 |
JarbasAI | In reply to @joergz:matrix.org Regardless of the poor quality for German - users of the Homeassistant forum also report this - it is not an option to use a RPI 5 as a server for STT. Including runtimes, the RPI 5 needs 17 to 18 seconds for approx. 3 seconds of speech to output the text. It made no difference whether I spoke German or English. However, the recognition rate for "my" English was 100 percent and for my German at most 5 percent. The Google Chromium service needs 0.9 seconds for this. Too bad. on a rpi4 with small model i was getting ~5seconds, dont think i tested with 3 seconds long audio, so assuming my tests had 1 second audio, sounds about right
a GPU (even a old crappy on) really helps here. if you need to run on a pi i would suggest Vosk
| 12:19:22 |
JarbasAI | for german you can also try the Nemo plugin from neon, but if i recall correctly users were not sure if it was better than whisper, it made different mistakes so was hard to judge. i think sgee_ did more testing around that | 12:20:40 |
goldyfruit | Yeah Nemo was the goto for german speaker based on sgee_ testing and the reason why I spawned the Nemo STT | 12:22:07 |
joergz | In reply to @goldyfruit:matrix.org Yeah Nemo was the goto for german speaker based on sgee_ testing and the reason why I spawned the Nemo STT But doesnt Nemo need a GPU? Where can I find an installation guide? | 12:23:12 |
goldyfruit | Nop | 12:23:21 |
goldyfruit | I running it via Docker, don't really have the steps for "normal" usage | 12:24:51 |
joergz | And I can download it from docker hub? | 12:26:04 |
goldyfruit | In reply to @joergz:matrix.org And I can download it from docker hub? https://github.com/OpenVoiceOS/ovos-docker-stt | 12:36:04 |
JarbasAI | thats what all public servers are using | 12:43:45 |
joergz | At least it is not a good plan to make STT local on a RPi even if it is a Pi 5, I guess. For me it was a test. I'll use public services again. Thanks! | 14:01:45 |
Mike | I haven’t found anything faster than Chromium or paid cloud services, but on GPU a large-v3 model does about 2-2.5s for most utterances. A bit slow, but still usable | 17:57:34 |
16 Jul 2024 |
Meliko | In reply to @builderjer:matrix.org I think if you use the adafruit 20w i2c amp, it has an i2c out that you could daisychain the respeaker to. Ovos-i2csound should autoconfigure both. I will try and test that theory soon How would i go around and daisychain those? Cuz i cant put them on top of eachother can I? | 08:05:04 |
Meliko | Fixed it | 09:22:31 |
Meliko | Can anyone tell me how i start ovos to listen to every time mycroft is heard, instead of just when i type ovos listen | 09:23:27 |
Meliko | In reply to @melikootje:matrix.org Can anyone tell me how i start ovos to listen to every time mycroft is heard, instead of just when i type ovos listen I ended up reinstalling it all, can anyone tell me, if i misconfigured the access key by typing accidental typos, is there any satellite command to redo the access key configuration? | 10:53:44 |
timonvanhasselt | This is maybe helpful (from the Hivemind documentation):
cat ~/.config/hivemind/_identity.json
You can maybe then edit the accesskey with nano ~/.config/hivemind/_identity.json
| 10:58:30 |
timonvanhasselt | * This is maybe helpful (from the Hivemind documentation):
cat ~/.config/hivemind/_identity.json
You can maybe then edit the accesskey with nano ~/.config/hivemind/_identity.json (with the right accesskey from the server)
| 11:00:59 |
JarbasAI | if ovos-listen works, i dont think he is using hivemind... that would require a standard OVOS to produce any output at all | 11:31:28 |
JarbasAI | In reply to @timonvanhasselt:matrix.org
This is maybe helpful (from the Hivemind documentation):
cat ~/.config/hivemind/_identity.json
You can maybe then edit the accesskey with nano ~/.config/hivemind/_identity.json (with the right accesskey from the server)
also, do not recommend editing the json manually, there is hivemind-client set-identity command to edit that file | 11:32:30 |
JarbasAI | this should probably go into the https://matrix.to/#/#OpenVoiceOS-Support:matrix.org channel too 😉 | 11:33:04 |
Meliko | In reply to @jarbasai:matrix.org this should probably go into the https://matrix.to/#/#OpenVoiceOS-Support:matrix.org channel too 😉 Ive continuer in support just now… its now just waiting for answers i guess | 11:39:21 |
timonvanhasselt | Oops sorry ☺️ | 13:12:15 |
JarbasAI | i mean, it is valid to edit the .json directly, just trying to avoid more json syntax erros from users 🙂 | 14:53:59 |
17 Jul 2024 |
Scienceguy | Hi everyone, I would like to point you to an opportunity for funding activities around the OVOS assistant. As some of you know, I work in a project called WASABI funded by the European Union program "Horizon Europe". We run two rounds of open calls to fund experiments involving digital assistants used in industry. Our first open call just launched this week (https://wasabiproject.eu/1st-open-call/). The rules for participation might seem a bit complex, but the consortium prepared a guideline and template, plus there will be an open webinar next week (that we also record for later). The call has a total of 1,250,000 € with a maximum funding per experiment of 125k for 12 months.
The rough idea is that WASABI funds small consortia to develop an OVOS skill (and if necessary a backend) and test it with an industrial end user. We will provide a basic Docker stack with a prepared configuration as well as an optional Android App to interact with it. IPR management must be outlined in each experiment to ensure the consortia can commercialize what they build (means we encourage commercialization of the experiment results). | 16:30:26 |
Mike | I’m assuming this is only open to EU citizens? | 22:41:58 |
18 Jul 2024 |
Scienceguy | Not exactly. You need an eligible organization to participate, i.e. typically in the EU. | 16:46:29 |
Scienceguy | That organization can hire you or subcontract. | 16:46:56 |
Scienceguy | The exact rules are in the call guidelines. | 16:47:12 |