What is the proper way to do local TTS?

[nick-pape]

Hi there, thanks for all the work on this!

So, I have everything setup and (mostly) working. I can easily ask Alexa to say something with:

action: notify.alexa_media
data:
  message: Hello!
  target:
    - media_player.nicholas_s_fire_tv_cube
  data:
    type: tts

She performs the TTS and speaks in her own voice. Wonderful.

Trying post: https://alexa.amazon.com/api/behaviors/preview : with uri: /api/behaviors/preview data {'behaviorId': 'PREVIEW', 'sequenceJson': '{"@type": "com.amazon.alexa.behaviors.model.Sequence", "startNode": {"@type": "com.amazon.alexa.behaviors.model.SerialNode", "nodesToExecute": [{"@type": "com.amazon.alexa.behaviors.model.OpaquePayloadOperationNode", "type": "Alexa.Speak", "operationPayload": {"deviceType": "A1VGB7MHSIEYFK", "deviceSerialNumber": "MY_DEVICE_SERIAL_NUMBER", "locale": "en-us", "customerId": "MY_CUSTOMER_ID", "textToSpeak": "Hello!", "target": {"customerId": "MY_CUSTOMER_ID", "devices": [{"deviceSerialNumber": "MY_DEVICE_SERIAL_NUMBER", "deviceTypeId": "A1VGB7MHSIEYFK"}]}, "skillId": "amzn1.ask.1p.saysomething"}}]}}', 'status': 'ENABLED'} query None
2025-02-15 15:18:32.160 DEBUG (MainThread) [alexapy.alexaapi] n******e@o*********m: POST: https://alexa.amazon.com/api/behaviors/preview returned 200:OK:application/octet-stream

 
However, I don't like Alexa's voice very much. She's made me too angry in the past. I want to hear from my TTS provider (OpenAI-TTS + Kokoro). I have this configured, it works fine for other media players, and the Assist chat.

However, when I target the Alexa Media Player device with:

action: tts.speak
target:
  entity_id: tts.openai_tts_bm_george_bm_lewis_2
data:
  cache: true
  media_player_entity_id: media_player.nicholas_s_fire_tv_cube
  message: Hello!

Alexa ends up saying: "Sorry, I'm having trouble accessing your Simon Says NA skill right now."

Logs indicate things are OK from the HA side:

Trying post: https://alexa.amazon.com/api/behaviors/preview : with uri: /api/behaviors/preview data {'behaviorId': 'PREVIEW', 'sequenceJson': '{"@type": "com.amazon.alexa.behaviors.model.Sequence", "startNode": {"@type": "com.amazon.alexa.behaviors.model.SerialNode", "nodesToExecute": [{"@type": "com.amazon.alexa.behaviors.model.OpaquePayloadOperationNode", "type": "Alexa.Speak", "operationPayload": {"deviceType": "A1VGB7MHSIEYFK", "deviceSerialNumber": "MY_DEVICE_SERIAL_NUMBER", "locale": "en-us", "customerId": "MY_CUSTOMER_ID", "textToSpeak": "<audio src=\'https://homeassistant.pape.house/local/alexa_tts/CxXeoz4J0sQmG171ZRwc7w_output.mp3\' />", "target": {"customerId": "MY_CUSTOMER_ID", "devices": [{"deviceSerialNumber": "MY_DEVICE_SERIAL_NUMBER", "deviceTypeId": "A1VGB7MHSIEYFK"}]}, "skillId": "amzn1.ask.1p.saysomething"}}]}}', 'status': 'ENABLED'} query None
2025-02-15 14:43:51.909 DEBUG (MainThread) [alexapy.alexaapi] n******e@o*********m: POST: https://alexa.amazon.com/api/behaviors/preview returned 200:OK:application/octet-stream

The only diff in the posts requests is the "textToSpeech". Basically, the audio tag or some other text.

A couple caveats.. 1) That URL is not publicly accessible, but it does have a valid SSL certificate. 2) It is accessible on the local network since I am using DNSMASQ. 3) I verified the file exists and is accessible by the Alexa device on that URL (used the Silk browser).

I feel like I am missing a config somewhere. However, looking at these, I couldn't tell what I am missing:

• Alexa Not Able to Stream TTS from Home Assistant #2739
• TTS Stopped Working #1683

I'm using latest HA and AMP versions.

Am I doing this wrong?

[nick-pape]

Okay, so I verified the audio files meets the requirements: https://developer.amazon.com/en-US/docs/alexa/alexa-presentation-language/apla-audio.html#audio-input-requirements

Checking on: https://developer.amazon.com/en-US/docs/alexa/alexa-presentation-language/apla-audio.html#url-requirements

I meet the first 2 (it is not a self-signed certificate, I just temporarily exposed HA to the internet, got a LetsEncrypt cert, then closed it back off -- I use VPN for outside access). But, I checked and my default NGINX proxy does not seem to be adding the CORS header. Checking on that now.

Edit: nope, the header didn't help.

Edit: Interesting. The audio file must be accessible to the public internet (i.e. Amazon servers), not just the Alexa device. Poking HA back through my firewall temporarily got it working. I'll need to figure out a better solution in the future to get this working. Might setup a VPN-connected serverless component to expose just that folder publicly.