Chirp 2

Specs & limits

• Official Speech-to-Text V2 model id is chirp_2 (underscore)
• GA, with regional GA expansion (us-central1/europe-west4/asia-southeast1) on 2025-01-27
• billed per audio

per Google official docs ↗

Use Chirp 2 in 30 seconds

OpenAI-compatible: swap the base_url, keep your SDK. POST /v1/audio/transcriptions

from openai import OpenAI

client = OpenAI(
    base_url="https://synthorai.io/v1",
    api_key="sk-syn-...",
)

resp = client.audio.transcriptions.create(
    model="chirp-2",
    file=open("meeting.mp3", "rb"),
    language="en",
)
print(resp.text)

import OpenAI from "openai";
import fs from "node:fs";

const client = new OpenAI({
  baseURL: "https://synthorai.io/v1",
  apiKey: "sk-syn-...",
});

const resp = await client.audio.transcriptions.create({
  model: "chirp-2",
  file: fs.createReadStream("meeting.mp3"),
});
console.log(resp.text);

curl https://synthorai.io/v1/audio/transcriptions \
  -H "Authorization: Bearer sk-syn-..." \
  -F model="chirp-2" \
  -F file=@meeting.mp3

package main

import (
	"context"
	"fmt"
	"os"

	"github.com/openai/openai-go/v3"
	"github.com/openai/openai-go/v3/option"
)

func main() {
	client := openai.NewClient(
		option.WithBaseURL("https://synthorai.io/v1"),
		option.WithAPIKey("sk-syn-..."),
	)
	f, _ := os.Open("meeting.mp3")
	resp, _ := client.Audio.Transcriptions.New(context.TODO(), openai.AudioTranscriptionNewParams{
		Model: "chirp-2",
		File:  f,
	})
	fmt.Println(resp.Text)
}

import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.audio.transcriptions.*;
import java.nio.file.Paths;

OpenAIClient client = OpenAIOkHttpClient.builder()
    .baseUrl("https://synthorai.io/v1")
    .apiKey("sk-syn-...")
    .build();

Transcription resp = client.audio().transcriptions().create(
    TranscriptionCreateParams.builder()
        .model("chirp-2")
        .file(Paths.get("meeting.mp3"))
        .build()).asTranscription();
System.out.println(resp.text());

About Chirp 2

• Google positions it as improving on the original Chirp in both accuracy and speed while adding word-level timestamps, model adaptation (phrase biasing), and speech translation.
• It offers automatic punctuation and capitalization, language-agnostic transcription, in which the model infers the prevalent spoken language and transcribes in it, and profanity filtering, and supports streaming, synchronous, and batch recognition for audio from under a minute up to eight hours, the longest batch ceiling of any Chirp model.
• Streaming covers sixteen locales, while batch carries the family's most extensive language support.
• Word-level timestamps are opt-in, and Google notes transcription quality and speed degrade slightly when they are on; the word-level confidence value it returns is documented as not a true confidence score.
• Adaptation takes simple words and phrases, but class tokens and custom classes are not supported.
• Speech translation runs asymmetric pairs, twenty-seven source languages into US English and eighteen targets out of it, and forced normalization and a denoiser round out the feature set.
• The one significant gap is speaker diarization, which Google lists as not supported; that capability is the reason to move to Chirp 3.
• Availability is limited to a small set of regions.
• Synthorai serves Chirp 2 transcription through its OpenAI-compatible audio endpoint.

FAQ

Related models

Compare