fal Assets is now live!
Docs
Log-inSign-up

nvidia/nemotron-asr-multilingual/asr

Nemotron-ASR-Streaming is a multi lingual, streaming Automatic Speech Recognition (ASR) engineered to deliver high-quality multi lingual transcription across both low-latency streaming and high-throughput batch workloads.
Inference
Commercial use
Streaming
Schema
LLMs
PlaygroundAPI

Table of contents

About

Run

1. Calling the API#

Install the client#

The client provides a convenient way to interact with the model API.

npm install --save @fal-ai/client

Setup your API Key#

Set FAL_KEY as an environment variable in your runtime.

export FAL_KEY="YOUR_API_KEY"

Submit a request#

The client API handles the API submit protocol. It will handle the request status updates and return the result when the request is completed.

import { fal } from "@fal-ai/client";

const result = await fal.subscribe("nvidia/nemotron-asr-multilingual/asr", {
  input: {
    audio_url: "https://v3b.fal.media/files/b/0a9c95c6/qxxx5skDQl8fPqbkjpxBc_speech.mp3"
  },
  logs: true,
  onQueueUpdate: (update) => {
    if (update.status === "IN_PROGRESS") {
      update.logs.map((log) => log.message).forEach(console.log);
    }
  },
});
console.log(result.data);
console.log(result.requestId);

Streaming#

This model supports streaming requests. You can stream data directly to the model and get the result in real-time.

import { fal } from "@fal-ai/client";

const stream = await fal.stream("nvidia/nemotron-asr-multilingual/asr", {
  input: {
    audio_url: "https://v3b.fal.media/files/b/0a9c95c6/qxxx5skDQl8fPqbkjpxBc_speech.mp3"
  }
});

for await (const event of stream) {
  console.log(event);
}

const result = await stream.done();

2. Authentication#

The API uses an API Key for authentication. It is recommended you set the FAL_KEY environment variable in your runtime when possible.

API Key#

In case your app is running in an environment where you cannot set environment variables, you can set the API Key manually as a client configuration.
import { fal } from "@fal-ai/client";

fal.config({
  credentials: "YOUR_FAL_KEY"
});

3. Queue#

Submit a request#

The client API provides a convenient way to submit requests to the model.

import { fal } from "@fal-ai/client";

const { request_id } = await fal.queue.submit("nvidia/nemotron-asr-multilingual/asr", {
  input: {
    audio_url: "https://v3b.fal.media/files/b/0a9c95c6/qxxx5skDQl8fPqbkjpxBc_speech.mp3"
  },
  webhookUrl: "https://optional.webhook.url/for/results",
});

Fetch request status#

You can fetch the status of a request to check if it is completed or still in progress.

import { fal } from "@fal-ai/client";

const status = await fal.queue.status("nvidia/nemotron-asr-multilingual/asr", {
  requestId: "764cabcf-b745-4b3e-ae38-1200304cf45b",
  logs: true,
});

Get the result#

Once the request is completed, you can fetch the result. See the Output Schema for the expected result format.

import { fal } from "@fal-ai/client";

const result = await fal.queue.result("nvidia/nemotron-asr-multilingual/asr", {
  requestId: "764cabcf-b745-4b3e-ae38-1200304cf45b"
});
console.log(result.data);
console.log(result.requestId);

4. Files#

Some attributes in the API accept file URLs as input. Whenever that's the case you can pass your own URL or a Base64 data URI.

Data URI (base64)#

You can pass a Base64 data URI as a file input. The API will handle the file decoding for you. Keep in mind that for large files, this alternative although convenient can impact the request performance.

Hosted files (URL)#

You can also pass your own URLs as long as they are publicly accessible. Be aware that some hosts might block cross-site requests, rate-limit, or consider the request as a bot.

Uploading files#

We provide a convenient file storage that allows you to upload files and use them in your requests. You can upload files using the client API and use the returned URL in your requests.

import { fal } from "@fal-ai/client";

const file = new File(["Hello, World!"], "hello.txt", { type: "text/plain" });
const url = await fal.storage.upload(file);

Read more about file handling in our file upload guide.

5. Schema#

Input#

audio_url string* required

URL of the audio file to transcribe.

language LanguageEnum

Target language for transcription (language-ID prompt). 'auto' lets the model detect the language. Default value: "auto"

Possible enum values: auto, en-US, en-GB, es-US, es-ES, de-DE, fr-FR, fr-CA, it-IT, ar-AR, ja-JP, ko-KR, pt-BR, pt-PT, ru-RU, hi-IN, zh-CN, vi-VN, he-IL, nl-NL, cs-CZ, da-DK, pl-PL, nn-NO, nb-NO, sv-SE, th-TH, tr-TR, bg-BG, el-GR, et-EE, fi-FI, hr-HR, hu-HU, lt-LT, lv-LV, ro-RO, sk-SK, uk-UA, mt-MT, sl-SI

acceleration AccelerationEnum

Controls the speed/accuracy trade-off. 'none' = best accuracy (1.12s chunks), 'regular' = balanced (0.56s chunks), 'high' = faster (0.32s chunks), 'full' = fastest (0.08s chunks). Default value: "regular"

Possible enum values: none, regular, high, full

{
  "audio_url": "https://v3b.fal.media/files/b/0a9c95c6/qxxx5skDQl8fPqbkjpxBc_speech.mp3",
  "language": "auto",
  "acceleration": "regular"
}

Output#

output string* required

The transcribed text from the audio.

partial boolean

True if this is an intermediate result during streaming.

{
  "output": "Actually, I'm second guessing myself a single coherent passage might be cleaner and more representative of real world speech performance, which is what benchmarking typically uses. I'll write an engaging flowing piece on something like a journey or a day in a tech forward city. That weaves in all the varied elements naturally, rather than splitting it into sections. I'm deciding to keep the targeted tricky sentences section after all. It'll directly test homophones and number heavy lines that are most prone to errors, so I'll include it as a concise, clearly marked addition at the end now. I'm drafting the full passage, starting with a detailed travel narrative that weaves in dates, times, currency, and specific details by the time we pulled into Waverley Station, my phone had buzzed forty seven times, mostly from doctor Amara Akon Reyes at Neurosynt Technologies fretting over our board presentation. We'd spent three weeks refining a transformer model with one point three billion parameters, and the improvements were stunning. Our word error rate plummeted from fourteen point two percent down to three point eight percent. There's something surreal about watching those number shift like that. They're not just metrics they're the payoff from countless late nights and too much coffee."
}

Other types#