Audio Support

Here we show how irouter makes audio queries as simple as text, image and PDF queries.

There are a limited amount of models that support audio input. Make sure to use an LLM that supports audio input, like google/gemini-2.5-flash.

To check which LLMs support audio input, check out the Model Overview.

Audio URLs

The simplest way to work with audio files in irouter is to pass an audio URL containing an .mp3 or .wav file:

from irouter import Call

model_name = "google/gemini-2.5-flash"
audio_url = "https://www.bird-sounds.net/birdmedia/241/860.mp3"

c = Call(model_name)
c([audio_url, "What bird is this?"], temperature=0)
# 'Based on the distinct "caw, caw" calls, this is an **American Crow**.'

Local Audio Files

You can also use local audio files by passing the filepath and an instruction as a list of strings:

from irouter import Call

model_name = "google/gemini-2.5-flash"
audio_path = "../assets/bottles.mp3"

c = Call(model_name)
c([audio_path, "What do you hear?"], temperature=0)
# 'I hear the sound of a glass bottle being opened and closed, along with the distinct clinking of glass.'

Chat with Audio

Use Chat for conversation tracking and token usage monitoring:

from irouter import Chat

chat = Chat(model_name)
chat([audio_path, "What do you hear?"])
print(chat.history)
print(chat.usage)