TTS API reading out emotion tags as well | Fish Audio | Page 1

jade hamlet Sep 25, 2025, 2:50 PM

#

Is it just me or your api doens't support emotion and empahsis tags?
These are working fine with your playground, but fail for API. Is there a captha you have built in to not allow this on API calls ?

tulip island Oct 3, 2025, 6:16 AM

#

Hi @rocky hedge were you able to solve that? It happens the same to me

fringe citrus Oct 5, 2025, 12:43 PM

#

tulip island Hi <@793276754331959326> were you able to solve that? It happens the same to me

Hi! Did you solve it?

granite night Oct 8, 2025, 8:06 PM

#

Hi! Same question here

sharp grail Oct 13, 2025, 3:18 PM

#

I find that if it is reading it out I have to run it multiple times. Eventually I can get it to work if I make small changes as I run the software. Sucks, but right now it is where the software is at. Learned to work around it but hope they improve that at some point in the future.

rocky hedge Oct 13, 2025, 9:27 PM

#

jade hamlet Is it just me or your api doens't support emotion and empahsis tags? These are ...

What @sharp grail said. You have to regenerate. My hit ratio on emotional emphasis tags is every 1 out 6 will use the tag correctly, rest go in the trash. After around 10 regenereations the sample returned will start getting severely mutated so wait, generate other sentances and come back later, repeat.

gusty canyon Oct 21, 2025, 9:24 PM

#

I came here for the same problem. I'm using almost exactly one of the samples from the SDK developer doc, and it's reading out loud the emotion tags. This doesn't happen in the playground, only when using the API:


session = Session("xxx")
request = TTSRequest( text="(happy) I'm excited to share this!" )

with open("test.mp3", "wb") as f:
    for chunk in session.tts(request):
        f.write(chunk)```

granite night Oct 21, 2025, 11:18 PM

#

Well, is it possible that it plays S1 in PG by default and v1.5 if you access through API? I don't see how to specify the voice model in API request using Python SDK.

tawdry juniper Oct 23, 2025, 3:03 AM

#

Hi! Make sure you are using S1 for emotion tag control

#

we are updating our docs, in the mean time please reference this repo for setting backend in python https://github.com/fishaudio/fish-audio-python

GitHub

GitHub - fishaudio/fish-audio-python

Contribute to fishaudio/fish-audio-python development by creating an account on GitHub.

tawdry juniper Oct 23, 2025, 3:36 AM

#

request = TTSRequest(
    reference_id="79973fbd9f374e21b29354e0477389b6",
    text="(happy) It brings me such joy to connect with you",
    format="mp3",
    # top_p=0.8,          # adjust as needed (0.0–1.0)
    # temperature=0.7,    # adjust as needed (0.0–2.0 typically)
)

with open("output1.mp3", "wb") as f:
    for chunk in session.tts(request, backend='s1'):
        f.write(chunk)

see this for python sdk

gusty canyon Oct 23, 2025, 3:36 PM

#

Can you provide an example of how to do that? I don't see any way to specify the model to use on either the Session or the TTSRequest (Session has some model related calls, but from what I can tell it looks like that is actually creating reference_ids, not models)

#TTS API reading out emotion tags as well