When making videos, making mini-games, or adding some sound to a page, you often get stuck on very small details.
The visuals are almost done, and only a drum beat is missing. You go to a sound library, but the ones with suitable rhythms have tones that are too heavy, while the ones with comfortable tones are mixed with vocals. After finally finding one, you cut it out and play it repeatedly, only to find an obvious pause between the end and the beginning.
The same goes for natural sounds. What you want is a stream and a few bird calls, but the material you find is mixed with loud wind noise, or suddenly has a piece of background music appear.
Many times, we do not need a complete song, but only “a small piece of sound that fits exactly here.”
In Ace Data Cloud's Suno API, sounds is the operation designed for this type of need. It generates sound material based on text descriptions, and can create both short sound effects that play once and sound clips for looping. For specific operations, refer to the Suno Audio Generation Integration Guide.

¶ First figure out whether you want one sound or a piece of music
Although both are submitted through the Suno audio API, different needs are suited to different operations.
A “pop” sound when a button is pressed, or the sound of wood being struck, is suitable for the one-shot mode of sounds. If you want a rhythm to play repeatedly, choose loop. If what you want is a long piece of music with a complete development process, then you should look at the standard generate.
This distinction is important.
This time, we created an intense looping rhythm, about six minutes of soothing music, and a natural soundscape of stream water and bird calls. Among them, the looping rhythm uses sounds; the latter two long audio clips use generate, and they cannot all be treated as generation results from sounds.
¶ Try loop with an intense drum beat
The requirements for this looping material are fairly straightforward: the rhythm should be more intense, with heavy drums and low frequencies, but it should not become a song with an intro, climax, and ending.
Because it needs to be played repeatedly.
If the model arranges a beautiful ending, it may sound good on its own, but it is easy to give itself away when connected back to the beginning. So in addition to the tone, the description should make the playback method clear: keep the rhythm stable, no fade-ins or fade-outs, no ending hit, and do not leave a blank section at the end.
The following clip is the sample generated this time using sounds with loop mode selected, approximately 20.8 seconds long.
The goal set this time was 140 BPM, using compact drum beats, hi-hats, and low-frequency synthesizer pulses to support the intensity. It is a creative target and does not mean that beat detection has already been performed.
Choosing loop also does not mean the file has already passed “seamless loop acceptance.”
This sample has not yet undergone seam listening tests or waveform inspection. Before actually putting it into a project, it is recommended to loop it continuously several times, paying particular attention to where the last beat connects back to the first beat: whether there is a pause, whether the low frequencies suddenly cut off, whether the reverb tail is cut off, and whether the rhythm rushes ahead or drags behind.
The player itself can also affect the experience. Even if the beginning and end of the audio are handled well, a gap may still occur after playback ends and the file is reloaded. Looping material and looping playback logic need to be checked together.

¶ The API is not complicated; the key is how to write sound
sounds does not use a separate /suno/sounds address, but is selected through action in the audio API. The request address is POST https://api.acedata.cloud/suno/audios, using Bearer Token authentication. See Suno Audios API Parameters for details.
The following is a looping rhythm request that can be used as a reference:
curl --request POST \
--url https://api.acedata.cloud/suno/audios \
--header "Authorization: Bearer YOUR_API_TOKEN" \
--header "Content-Type: application/json" \
--data '{
"action": "sounds",
"model": "chirp-v6",
"sound": "Intense industrial percussion loop at 140 BPM. Punchy kick, tight snare, fast hi-hats and gritty D minor bass pulses. Stable tempo and energy. Seamless cyclic boundary, complete bars. No intro, outro, fades, ending hit, silence or vocals.",
"sound_type": "loop",
"bpm": 140,
"key": "Dm",
"audio_format": "mp3",
"async": true
}'
Looking at this request piece by piece, the part that really requires thought is mainly still sound.
Writing only “make it more intense” expresses an emotion. Adding “heavy kick, tight snare, fast hi-hats” starts to get closer to the sound itself. Then stating “keep it stable, with no intro or outro, for continuous looping” lets the model know how this material is intended to be used.
We recommend making the description specific, rather than continuously piling on adjectives such as “stunning,” “ultimate,” and “epic.”
For example, a button sound effect can be described as:
A short, soft wooden tap, with a clean, close sound, fast decay, no vocals or background music.
Likewise, for natural sounds, “there is flowing water and birds in a forest” is still relatively broad. Changing it to “the stream is closer, the bird calls are farther away, leave quiet space between calls, and do not have continuous dense chirping” makes the spatial relationship and density much clearer.
The description does not need to be very long. In this test, a sound field exceeding 500 characters was rejected by the API, and it was only submitted successfully after being shortened. During integration, you can first keep it within 500 characters and put the most important sound characteristics at the beginning; the length limit should still be based on actual validation and the latest documentation.
¶ These parameters are enough to get started
The required fields for sounds are action, sound, and sound_type. The remaining sound-effect parameters are used to supplement the model, tempo, key, and output format.
| Parameter | How to fill it | When to use it |
|---|---|---|
action |
Fixed as "sounds" |
Select the sound asset generation operation |
sound |
Text description | Describe the sound source, timbre, sense of space, and playback requirements |
sound_type |
"one-shot" or "loop" |
Distinguish between one-shot sound effects and loop assets |
model |
This article uses "chirp-v6" |
Explicitly specify the generation model |
bpm |
An integer from 1 to 300 | Fill in when there is a clear tempo requirement |
key |
For example, "Dm" or "C" |
Fill in when there is a key requirement |
audio_format |
Currently "mp3" |
Specify the audio format |
async |
Set to true |
Obtain a task ID after submission, then query the result |
callback_url |
Optional result receiving address | Fill in when a server-side callback is needed |
Among them, the range of bpm, the valid values of sound_type, and the current MP3 format limitation can all be found on the API parameter page.
If you are only generating running water, wind, or a single knock, there is no need to force-fill BPM and key just for “complete parameters.”
There is another easy pitfall: /suno/audios is a shared entry point for multiple operations. Fields appearing in the overall parameter table do not mean they are valid for every operation. For example, the documentation states that duration is currently used for generate; you cannot therefore assume that passing 360 to sounds will give you a six-minute sound effect.
¶ For six minutes of creek water and birdsong, use a different approach
Beyond short loops, we also tried a longer natural soundscape.
This time, the goal was not to present “nature is very lively,” but the feeling of sitting by a creek for a while: the flowing water is continuously present, birdsong sounds occasionally, there are variations in distance, but they do not compete with each other.
In the sound design, the creek serves as the base layer, while birdsong serves as decoration. Only a little rustling leaves and wind are retained, with no thunder or sudden rain arranged, and no instruments actively added.
Below is an approximately six-minute sample of Forest Creek.
This needs to be made clear: it uses generate, not sounds, and it is a synthesized soundscape, not a field recording.
To try longer continuous content, this test used chirp-v5-5 custom generation and set duration to 360. The documentation also reminds us that duration is only a generation target, not an exact guarantee; the actual duration in the returned result should be checked in the end.
Likewise, writing “no music” cannot replace listening-based acceptance. This sample has not yet completed full manual listening; the fidelity of the running water, birdsong, and no-background-music requirement needs to be confirmed before use.
If a project has very strict requirements for pure environmental sound, you can first verify the sound quality with short segments, then consider a longer production workflow. There is no need to force all requirements into a six-minute audio clip from the beginning.

¶ Soothing music should not be confused with environmental sound either
We also created an approximately six-minute piece called Slowly Settling Into Calm.
Its creative direction is soft piano, warm ambient padding, and slowly changing harmony. Unlike creek water and birdsong, this is a piece of music; unlike intense loops, it allows for a slow entry, development, and gradual ending.
This piece also uses generate and is not a sounds sample. It is suitable as a creative attempt in the direction of supporting relaxation, and should not be presented as audio with definite therapeutic effects, nor does it make the promise that “a certain frequency can cure illness.”
When these different needs are considered together, the API selection becomes relatively clear:
If you need a sound event, or a piece of sound material for looping, look at sounds first; if you need musical content that unfolds over a long period, then look at generate.
¶ After obtaining the task_id, do not submit again immediately
The example uses async: true. After submission, you first receive a task ID, then query the result through the task API, instead of continuously waiting on the same generation request.
curl --request POST \
--url https://api.acedata.cloud/suno/tasks \
--header "Authorization: Bearer YOUR_API_TOKEN" \
--header "Content-Type: application/json" \
--data '{
"action": "retrieve",
"id": "替换为提交时返回的task_id"
}'
During integration, it is recommended to save the task ID together with the request record. As long as the original task is still running, continue querying that task. Do not send another generation request just because the file is not available yet.
After obtaining the final audio, do not only check whether “there is a URL.” You should also check whether the task succeeded, whether the actual duration is suitable, and then evaluate the sound content according to its intended use.
For short sound effects, check the attack and decay; for loop assets, listen for the seam; for long audio, listen for abrupt transitions, unexpected vocals, or unwanted background music.
The generation API solves the problem of “first creating a candidate asset.” To make it truly suitable for your page, video, or game, this final step of listening and adjustment is still worth setting aside time for.
The three illustrations in this article are original generated images created for this test. The audio is generated sample content and has not yet completed full manual listening review or professional audio quality inspection. The blog uses native <audio> tags for embedding; the publishing platform needs to allow these tags to be retained for readers to play them directly.

