[renamed] Miscs, Discussions, ElevenLabs, PSRAM, Updates & NEWS

[huybachkhoa2702]

Hi I want to use this code for the built-in PDM micro on some ESP32 S3 boards from Xiao or DFrobot... Can you help me how to calibrate it? Thanks

[kaloprojects]

No, unfortunatly not, I don't have this micro, neither any datasheets.
I am pretty sure it could work too .. but i have no details about all correct entries for the I2S structures for your micro, neither can test. The music (HW & protocol details) plays here:

i2s_std_config_t std_cfg = ....
{ .. // Datasheet INMP441: Microphone uses PHILIPS format (bc. Data signal has a 1-bit shift AND signal WS is NOT pulse lasting)
.slot_cfg = I2S_STD_PHILIPS_SLOT_DEFAULT_CONFIG( I2S_DATA_BIT_WIDTH_16BIT, I2S_SLOT_MODE_MONO ),
...
.bit_shift = true, // important ! (that's the reason why we need the PHILIPS macro, not the MSB macro) !
.msb_right = false, // because WIDTH_16BIT used
// Update: ESP32-S3 (!) does NOT have a final '.msb_right' ... instead it has 3 other elements at eof structure:
// .left_align = true, .big_endian = false, .bit_order_lsb = false,
}

so best would be you check/research about all I2S property details for this micro (also using I2S Espressif links, that's how i did on my micro).. and insert & test those values into the .ino header. Sorry for not having better answer ;)

[RWBRWB]

Chat GPT is pretty good at alot of coding task so I would give that a try to integrate what your wanting @huybachkhoa2702

[palashis92]

GPT Is Good But Grok and Claude is Invincible.

[palashis92]

Hey @kaloprojects , how’s it going?
Do you have any plans to integrate ElevenLabs TTS?

Currently, ElevenLabs doesn’t provide any streaming links, so we can’t connect it using the Audio.h library with a POST method. Maybe we could request the ElevenLabs team to add something like that?

By the way, I really appreciate the hard work you put into getting OpenAI TTS working on the ESP32 without PSRAM — that’s impressive!
But unfortunately, I’m currently using a combo of Scribe, LLaMA, and gTTS. It feels like a rocket on the ESP32 without PSRAM — response time is just 3 to 3.5 seconds max.

OpenAI TTS really frustrated me — the TTS alone takes 5 seconds, and the LLM adds another 2–3 seconds.
I’m hoping ElevenLabs can offer faster TTS responses.

[RWBRWB]

GPT Is Good But Grok and Claude is Invincible.

I just got Claude Code to test it out. Pretty cool how it works inside VS Code.

Haven't tried Grok for coding.

[palashis92]

GPT Is Good But Grok and Claude is Invincible.

I just got Claude Code to test it out. Pretty cool how it works inside VS Code.

Haven't tried Grok for coding.

Try Grok — it's amazing for coding!
And if you're into hardcore development, you should definitely check out Cursor.ai or the Replit AI Agent — both are powerful tools for serious coders.

[RWBRWB]

Do you have any plans to integrate ElevenLabs TTS?

I think they have some cool features but cost is 3x more expensive which sucks. I find the TTS-1 to be the most expensive part of running this chatbot so would hate to 3x that expense but Elevenlabs does offer some unique voice cloning that would make things alittle more fun.

[kaloprojects]

Hi @palashis92 : i am doing well, thanks ;)
The problem on TTS Elevenlabs are also the costs, pretty expensive i feel (the STT is cheap, the TTS is not), currently ~20K characters in free abo, and 60K on 5$ abo). You can reach both limits easily on few chats (Open AI new TTS is powerful and much less expansive). And btw .. Open AI did some nice enhancements, with new models (beyond tts-1), there are new featues available, e.g. https://www.openai.fm/

But yes, i will work on a Elevenlabs TTS implementation (in future).. just not in the next weeks, because i have to focus on my next upgrade (which will have some more nice AI features, you will see ;). So i am busy in moment.

kalo

[kaloprojects]

funny .. RWBRWB and i wrote in same minute LOL 😆

[palashis92]

funny .. RWBRWB and i wrote in same minute LOL 😆

Look at my and @RWBRWB username it's More Funny.

[palashis92]

Hi @palashis92 : i am doing well, thanks ;) The problem on TTS Elevenlabs are also the costs, pretty expensive i feel (the STT is cheap, the TTS is not), currently ~20K characters in free abo, and 60K on 5$ abo). You can reach both limits easily on few chats (Open AI new TTS is powerful and much less expansive). And btw .. Open AI did some nice enhancements, with new models (beyond tts-1), there are new featues available, e.g. https://www.openai.fm/

But yes, i will work on a Elevenlabs TTS implementation (in future).. just not in the next weeks, because i have to focus on my next upgrade (which will have some more nice AI features, you will see ;). So i am busy in moment.

kalo

Yeah Openai did great enhancements but When i chnage TTS-1 To GPT-4o-mini-tts It Sucks ..No audio as Output.

[kaloprojects]

Yeah Openai did great enhancements but When i chnage TTS-1 To GPT-4o-mini-tts It Sucks ..No audio as Output.

PSRAM needed for this feature (older AUDIO.H don't support) but i think about a self made patch back to enable this also for older ESP32, we will see ;)

[RWBRWB]

It works for me ...

Yes, because you have PSRAM and can use the latest AUDIO.H version ;) .. that's what i meant ;)

kalo wrote: ... SORRY .. i wrote in wrong post, omg. .. must leave anyhow now .. have a great day !, bye

[palashis92]

Yeah Openai did great enhancements but When i chnage TTS-1 To GPT-4o-mini-tts It Sucks ..No audio as Output.

PSRAM needed for this feature (older AUDIO.H don't support) but i think about a self made patch back to enable this also for older ESP32, we will see ;)

No Need i will Upgrade To ESP 32 S3. But let me know How much time it took to responce With PSRAM?
and One Feature Request Please add an Wake Word.. I Really Need it for my Alexatron.