The video is almost finished, and the subtitles are all matched up, but somehow it gets stuck on the voiceover.

Switch to another voice, and it sounds too much like an advertisement; switch again, and the words are pronounced clearly, but it sounds like someone reading a notice. Try to find something more relaxed, and it ends up being too relaxed. After browsing the audition list for ages, you still do not know whom to choose.

Many times, it is not that the copy is bad, but that the voice and the visuals do not match.

So this article will not pile on the parameters first. We found some distinctive voices in the Fish Audio community, wrote original short sentences that match their respective expressive directions, and then actually generated audition clips through calls. You can listen first, then think about this: what kind of person does your own video actually need to tell its story?

A creator agonizing over choosing a voice, with cassette tapes, scripts, and headphones laid out in front of them

When choosing a voice, do not just look at the name

Labels such as “gentle female voice” and “magnetic male voice” all sound nice, but they may not necessarily fit once placed into a video.

When talking about everyday life, sounding too formal can create a sense of distance; when doing historical narration, being overly lively can easily break the immersion. Whether a voice sounds good is one thing; whether you are willing to listen to it finish telling this content is another.

You might as well first ask yourself one question: who is saying these words to whom?

Is it friends chatting, someone telling a story, or a character on stage? Once you figure this out, you will not need to browse the voice list from beginning to end.

Fish Audio can generate speech from text and can also call existing voices. Its Chinese official website introduces capabilities such as multilingual support, voice cloning, and emotional expression; for the specific results, you still need to test them with your own script. Fish Audio Chinese Official Website

For everyday creation, there are two points worth noting: one is that there are many voice choices, so you do not have to keep circling around just a few announcer-style voices; the other is that you can repeatedly call the same voice ID, which is more convenient when revising scripts or adding lines later.

If you have voice material that has already been authorized, you can also create a one-time reference audio synthesis without first creating a long-term voice. Here, you need to distinguish clearly: the voice model determines “who it sounds like is speaking,” while the speech model determines “which synthesis capability is used to generate it”; they are not the same thing.

When connecting through Ace Data Cloud, reference_id specifies the voice, while model in the HTTP request header specifies the speech model. The current API documentation lists s1, s2-pro, and s2.1-pro, and supports options such as MP3, WAV, and speed adjustment. Parameter Description

If you want to integrate it into your own tool, the code can be very short

What is demonstrated below is the Fish TTS API from Ace Data Cloud, not the official direct Fish API. Use Ace Data Cloud's API Token; the example reads environment variables and does not write secrets into the code.

import os
import requests

# 换成你已确认使用权限的音色 ID
voice_id = os.environ["FISH_VOICE_ID"]

response = requests.post(
    "https://api.acedata.cloud/fish/tts",
    headers={
        "Authorization": f"Bearer {os.environ['ACEDATACLOUD_API_TOKEN']}",
        "Content-Type": "application/json",
        "model": "s2-pro",
    },
    json={
        "text": "别急着挑最好听的，先挑一个适合这段故事的。",
        "reference_id": voice_id,
        "format": "mp3",
        "prosody": {"speed": 1.0, "volume": 0},
    },
    timeout=180,
)
response.raise_for_status()
result = response.json()
if "audio_url" not in result:
    raise RuntimeError(f"未取得音频地址：{result}")
print(result["audio_url"])

First install requests, then configure ACEDATACLOUD_API_TOKEN and FISH_VOICE_ID, and you can test it once with a short sentence. The obtained audio_url is an audio address, not the audio file itself.

Do not generate the entire piece right away. Choose a section of the script with long sentences, pauses, and emotional changes, then continue after testing out an appropriate speed. For long texts or batch tasks, you can use callback_url and task queries according to the documentation to avoid making the page wait continuously.

How many of the voices below do you recognize?

Let us make this clear first: the following 20 audio clips are all AI-generated audition clips created this time by calling community voices through Ace Data Cloud Fish TTS. They are not recordings by the individuals themselves, nor are they original lines or original voice acting. The copy is original for this occasion, and the titles retain or explain the community voice names; they do not represent the participation, authorization, or endorsement of the individuals, voice actors, or related organizations.

We selected a mix of Chinese conversational, narration, and distinctive character voices; this is not a real-time popularity ranking. Each clip uses a different short scenario sentence, mainly to help convey the direction of expression, rather than serving as a comparative evaluation using a unified script. Anon Chihaya uses Japanese, while the rest are Chinese; results in different languages cannot be directly compared.

The audio links have been checked for accessibility, but the listening experience has not yet been reviewed clip by clip. The descriptions below explain the intended design of the copy and do not constitute a guarantee of how the generated results will sound.

Successful generation does not mean authorization for use has been obtained. Public community models do not mean they can be used for commercial advertisements, impersonating the person, or false endorsements; before formal publication, display, or reuse, you still need to confirm the voice, the rights to the person's voice, and character-related rights. This draft remains unpublished.

Original characters with different voices taking turns standing before the microphone, forming a lively little theater of voices

Crayon Shin-chan｜That familiar mischievous spirit

Let us start with the childhood channel. For this voice, we wrote a mischievous excuse revolving around little cookies, hoping to show the contrast of a child seriously looking for an excuse. Besides the pitch, you can also listen to the pauses and sentence endings.

Mom, I really did not slack off today! I was seriously researching why, as soon as a little cookie is opened, it runs into my tummy all by itself. Hehe, give me another one!

Sun Wukong｜The character feel of “This Old Monkey”

Use scouting the way ahead, protecting his master, and the Golden Cudgel to string together a short piece of original character dialogue, leaving some room for boldness and dramatic tension. You can compare it with ordinary narration and listen to where the sense of character comes from when it is still just speaking.

Master, do not worry! The mountains ahead are high and the roads are treacherous; let this old monkey go and investigate first. If any demons block the way, this Golden Cudgel of mine is not for use as a walking stick!

SpongeBob SquarePants｜I want to go jellyfishing today too

Designed as an excited invitation to head out, with copy that is bright, lively, and full of anticipation. If the visuals themselves are already very lively, whether the voice-over should still be this full can be decided after trying it in.

Patrick, wake up! The sea is sparkling today, and the jellyfish must want to come out and play too! I even have the net ready—I’m just waiting for you, my best partner!

Patrick Star｜Let’s eat a Krabby Patty first, then think

Connect it with the previous segment as a short dialogue: one is in a hurry to go out, while the other is thinking about eating first. Here, use logic that is half a beat slow and takes a little turn to show a silly light-comedy direction.

SpongeBob, I thought of a particularly clever way. Let’s eat a Krabby Patty first, then think about where to play. That way, if we can’t think of anything, at least we won’t go hungry.

Lazy Goat｜Let me lie down for a while first

The cake mountain in a dream and the nap one cannot wake up from set the direction of being greedy and sleepy for this copy. If you want to make more relaxed content, there is no need to rush every sentence forward.

Don’t make me get up, I just dreamed of a little mountain made of cake. Once I finish eating the strawberry at the top, I’ll get up and help. Really, just sleep a little longer.

Emperor from Empresses in the Palace｜A scene emerges from a single line

This is an original palace monologue and does not quote dialogue from the series. Use the scene of putting down a teacup and hesitating to speak to make room for the character’s sense of status and implication.

There is no need to refill this tea. I wish for some quiet for a moment. I have remembered all that you just said. It is only that, in this palace, what one hears may not be true, and what one sees may not be the whole picture.

Uma Musume｜Look in the direction of energy

Choose a scene of running one more lap before training ends, using requests, persistence, and anticipation to drive the dialogue. If you want to find more spirited character expression, you can start from this kind of copy direction instead of merely searching for “sweet female voice.”

Trainer, the wind is just right today! For the final lap, I want to try once more. Even if I am still a little short right now, I will catch up step by step. Please watch me dash across the finish line!

Anon Chihaya｜Switch languages and listen for the rehearsal feel

A short Japanese band rehearsal line was written for this voice, meant to convey the feeling of being a little nervous while wanting to pull her partners into trying one more time. This is not a Chinese pronunciation test, nor is it an original recording from the work.

Hey, how about we play through today’s practice just one more time? I think it’ll definitely sound better than before. I’m a little nervous, but with everyone, it’ll be okay!

Chinese gist: How about we play through today’s practice one more time? I think it will definitely be better than before. Although I’m a little nervous, it will be fine when I’m with everyone!

You don't necessarily have to recognize the name to find a voice you like

Character voices are easy to remember, but when choosing voice-overs for your own videos, it often still comes down to “does it sound natural” and “does it become tiring to listen to for a long time.” The following examples can be listened to from a different angle.

AD Senior｜A short line, listen to the tone first

The script is set as a team-leading reminder in a game, aiming for a bit of sharpness and playfulness, but without relying on attacking others to create momentum. Short sentences make it easier to feel the tone; long videos still require separate testing for continuous narration.

Don't just keep saying you're no good—finish this round first. It's okay to move a little slower, but don't let your brain go offline too. Come on, stay close to me. We'll play this wave seriously.

Jialan 3.0｜Like telling a friend about something

The little conflict of “being scared, but still wanting to play” is very suitable for writing like a conversation between friends. Repetition, hesitation, and adding a condition at the last minute can all become part of a spoken script; there is no need to make every sentence read like an instruction manual.

I'm telling you, I really don't dare play that game alone. But if you're with me, I feel like maybe it could work. But let's agree first—if a door makes a sound, you can't run before me!

Female College Student｜The library, dinner, and a cup of milk tea

This is the name of a community voice, not an identity verification. An ordinary campus daily-life passage was chosen, letting a study plan and the little thoughts of getting distracted squeeze into the same paragraph, to test the direction of natural, everyday narration.

I finally got a window seat in the library today. The sunlight is especially nice, and before I even turned two pages, I had already started thinking about what to eat for dinner. Forget it, I'll finish memorizing this chapter first, then reward myself with a cup of milk tea.

Gentle and Pleasant Female Voice｜Not every sentence needs to rush to speak

This script is quiet companionship, not urging. If you want to make bedtime stories or soothing narration, you can focus on testing long sentences, pauses, and sentence endings in this kind of direction.

If you're a little tired today, put down what you're doing for now. The wind outside the window is gentle, and the tea is still warm. You don't need to rush to become better; at this moment, just breathe slowly.

Ding Zhen｜Talk about the scenery, rather than reciting an introduction

Original short scenic lines about snow-capped mountains, dew, and a small path were chosen, with the intended expression being gentle and conversational. This is an AI synthesis demonstration of a community model with the same name, not a statement, travel recommendation, or endorsement by the person himself.

When I went out this morning, there was still a little snow on the mountaintop, and the dew on the grass was shining. Walk slowly along this little path; there's no need to rush. The wind will carry the sounds from far away over here.

Cai Xukun｜A lighter expression in a young male voice

A fictional practice-room monologue was written, using small movements and music to bring the topic toward a stage atmosphere. This is not the person's real experience, interview, or statement, and is only used for a community voice synthesis demonstration.

The lights in the practice room are still on. I want to do that move from just now one more time. Sometimes progress is hidden in these little places. When the music starts, all the hard work will have an echo.

Sun Xiaochuan 258｜A more casual chatting direction

A short self-deprecating passage was written using staying up late, gaming, and takeout that has gone cold, looking for a casual chatting rhythm rather than formal broadcasting. The script is fictional, not the person's experience or real statement.

I said I'd go to bed early today, but then I looked at the time, and it's this late again. I didn't win the game, but the takeout got cold pretty fast. Fine, I'll eat first. After I'm full, I'll figure out whose fault that was just now.

For an explanatory feel or a sense of storytelling, there are other options too

Lei Jun｜Speaking along with the audience below the stage

Use a hypothetical question about a chair to write an explanatory practice, trying a sense of interaction that “asks first, then explains.” It does not involve any real product launches, price or performance promises, and does not represent his personal views or endorsement.

Everyone, think about it: for a chair, what is the most important thing? It is not how complicated it looks, but whether, at the moment you sit down after a busy day, you can truly relax. Doing the small things well is also a kind of skill.

Zheng Xiangzhou｜A more formal business explanation

The original copy goes from goals and division of responsibilities to results, giving the formal explanation a clear structure. It can be compared with conversational passages to feel the sense of distance brought by changes in formality. The content is not his personal view or a recording of his lecture.

When building a team, do not rush to talk about scale first. Explain the goals clearly, put the division of responsibilities into place, and then let everyone know what relationship their work has with the results. Being aligned in direction is more important than appearing busy on the surface.

Documentary narration｜Giving scenic writing a little more weight

The community model was originally named “CCTV dubbing.” Here, “documentary narration” is used to describe the direction of expression and does not indicate any cooperation with or authorization from CCTV. This time, an original short piece about a river valley at dawn is used; the arrangement of long sentences is worth noticing.

At dawn, the first ray of sunlight crosses the ridge, and the river valley gradually wakes up. The water flows around the rocks and through the village. Those seemingly ordinary days slowly leave their traces in the sound of the flowing water.

Grand promotional film｜When the opening needs to be a little more formal

Use city lights and the lives of ordinary people to write short opening lines, without adding product promises. This kind of expression does not have to be used throughout the entire video; you can place it in the first few opening lines and see whether it feels too full.

From the first light at dawn to footsteps that have still not stopped late at night, the warmth of a city is hidden in the hearts of everyone who lives earnestly. Let the ordinary be seen, and let every effort have an echo.

Black Hand｜Adding a little sense of plot to a monologue

A fictional monologue about a rainy night, an old letter, and leaving was written, with a restrained, serious, and story-driven direction. This kind of copy is more suitable for character practice and does not need to be forced onto every kind of narration script.

When the rain stopped, he put that letter back into his pocket. The light on the street corner was still on, but no one was waiting for his explanation anymore. Some roads, once you turn away, can only be walked alone until dawn.

After listening, do not rush to decide—put it back into your own footage

A script passes through a fish-shaped radio and becomes a strip of sound paper; a creator wearing headphones checks it at an editing desk

A voice you like during the audition may be another matter once it is put back into your own video.

You can first keep two or three candidates, generate them using the same script, and then listen with the same footage. Do not compare only the first sentence; also listen to whether long sentences feel rushed, whether sentence endings are held properly, and whether it is still clear after background music is added.

If it still does not feel right, revise the script first. Use less written language, split up long sentences, and leave a pause at transitions. Sometimes, changing ten voices is not as effective as revising a sentence into the way you would actually say it in everyday life.

Fish offers more voice options; whether the final result is suitable still depends on your ears. If it can make viewers forget “this part is voice-over” and be willing to keep listening to the content, that is already very good.

If you want to keep trying, you can start here

The illustrations in this article are concept illustrations created with AI assistance, not product interface screenshots. The 20 audio clips are AI-generated demonstrations created by calling Fish community voices using original copy written for this use. They are not recordings of the individuals themselves or of the original works, nor do they represent authorization, participation, or endorsement by the relevant individuals, voice actors, or organizations. Before formal publication, please complete a clip-by-clip listening review, check frontend player playback, and confirm that the authorization scope for the voices, characters, and audio presentation covers their subsequent intended use.