by Fish Audio·Released 13 October 2026
#4 for realistic voice · Drama 3 performs a script the way you direct it, reading plain-language stage directions in square brackets before each line, with sighs, pauses and whispered words placed exactly where you write them
You pay for what each request uses. No subscription, no minimum.
80 % offFish Audio has cut its price. Every figure here is the new one.
One paragraph read aloud, the ordinary use
Every rate
| Meter | ↑ you send | ↓ comes back |
|---|---|---|
| UTF-8 bytes | 665,67 DA |
per 1M UTF-8 bytes
One company runs this model. OpenDunes sends every request to it. DA per 1M UTF-8 bytes: ↑ what you send.
| Provider | ↑ Sent | Latency | Uptime |
|---|---|---|---|
| Fish Audio80 % off | … | … |
You pay the price of the provider that served your request. Prices read 15:18 · speed over the last 30 minutes · uptime over 3 days.
Swap the base URL and the key. Your OpenAI code runs unchanged.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://opendunes.com/v1",
api_key=os.environ["OPENDUNES_API_KEY"],
)
with client.audio.speech.with_streaming_response.create(
model="fish-audio/drama-3",
input="مرحبا بكم في متجرنا. التوصيل مجاني في وهران.",
voice="alloy",
) as resp:
resp.stream_to_file("speech.mp3")Runs this request once on your balance.
About 0,40 DA per paragraph read aloud. Was 2,00 DA. ↑ 665,67 DA per 1M UTF-8 bytes. Arabic text takes about two bytes a character.
No. You pay per use from your balance, with no subscription. Try it in Chat at the same price.
#4 for realistic voice · Drama 3 performs a script the way you direct it, reading plain-language stage directions in square brackets before each line, with sighs, pauses and whispered words placed exactly where you write them.
Deposit dinars with your Edahabia or CIB card. Every model, Drama 3 included, is paid from that balance. No foreign card, no conversion.
Fish Audio names no languages for Drama 3. Arabic, French and Darja: not stated.
Voices: not stated. Cloning from a sample: yes.
Fish Audio. Released 13 October 2026.
Transcribes audio with automatic language detection and, when alignment details are asked for, timestamped word-level segments; the general Fish Audio transcription model, where Transcribe 1 Pro adds speaker labels
Transcribes interviews, meetings and podcasts, marking each speaker inline, keeping emotion and vocal cues such as laughter, detecting the language and returning timestamped words; the Pro tier of Transcribe 1
#5 for realistic voice · Fish Audio's production-grade successor to S2 Pro, for multilingual voice apps, narration and dialogue, taking open-ended direction on style and emotion in plain words and cloning a voice from a reference
Fish Audio's multilingual voice model for applications that need a wide emotional range, where the speaking style is steered with short notes in parentheses inside the text rather than written instructions
From what its maker and two benchmark sites publish, and where it stands on OpenDunes. Read 7 Nov 2026.
Clones a voice
Speaks in a voice you record.
Act out a scene between several speakers, each turn tagged, in one request
Pin how a brand name or a surname is said with phoneme tags
Write delivery notes in French for a script that is spoken in English