OpenAI Launches New Voice to Text API

OpenAI announced the official launch of ChatGPT and Whisper voice to text API for business users,and gave a series of cases where business partners have been online.

In short,after the launch of ChatGPT chat robot to the public in November last year,OpenAI now offers paid access products to enterprises that intend to provide the same AIGC service in their own APP or products.The company said that the API provided this time is based on the GPT 3.5 model of the same model of ChatGPT.More importantly,after last December,the company has successfully reduced the cost of ChatGPT by 90%.

Therefore,this model named”gpt-3.5-turbo”is priced at$0.002/1000 tokens.According to the OpenAI official website,token can be understood as an unstructured word,while 1000 tokens correspond to about 750 words.This price is also 90%cheaper than the current GPT 3.5 model.

Several commercial applications have become early users of ChatGPT API.Photo and short video social platform Snap launched a customizable chat robot called”My AI”this week for subscription users.The Quizlet online learning platform,which has 60 million student users,provides tutoring robots that can test students.The Shopify platform,which many Chinese retailers will use to sell goods at sea,has also launched chat robot shopping guides.Within a few months,ChatGPT quickly found its place in the global software ecosystem.

In addition to chat robots,OpenAI also put forward a commercial voice to text solution on Wednesday.The company first launched the Whisper voice to text model in September last year,and today it officially offered the API interface for business partners to pay for access,with a price of$0.006 per minute.

According to the company,Whisper API supports the transcription and translation of voice files,and supports dozens of languages including English,Chinese,Arabic,Japanese,German,Spanish,etc.

However,it is worth noting that the product description document of OpenAI also shows that the word error rate of Whisper large-v2 model in recognizing English,Italian and German can be controlled below 5%in the FLEURS data set test commonly used in the industry.