© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
Weeks after it stunned the world with its Her-like audio interface, OpenAI has finally begun rolling out its advanced Voice Mode. As of now, the company has begun rolling out it to a small number ChatGPT Plus subscribers. When it was introduced at the Spring Update event along with GPT-40, OpenAI drew flak as the Voice Mode bore a striking resemblance to Hollywood actor Scarlett Johansson who famously voiced the AI system in filmmaker Spike Jonze’s ‘Her’. The advanced mode was to be released in alpha sometime in June, however, OpenAI delayed the rollout by a month.
The new Voice Mode is simply not ChatGPT with voice. During the event, staff from OpenAI demonstrated how it can hold conversations like humans, be part of conversations in a group setting, and how it can adjust itself to the kind of conversations around it. The delay in the launch of the advanced mode was due to OpenAI’s ongoing work towards improving the model especially to hone its ability to detect and refuse certain content.
Reportedly, OpenAI has tested the voice model’s abilities with over 100 external experts or red teamers. In May, when the company showcased the voice model for the first time it attracted criticism from certain sections owing to its eerie similarity to the voice of Johansson. Following the demo, OpenAI ran into controversy after the actress said that she had asked CEO Sam Altman not to use her voice for any OpenAI models. She later sought legal counsel, however, OpenAI denied that they used Johansson’s voice. However, the company later removed the voice.
There is a Voice Mode currently available on ChatGPT, however, it is radically different from the advanced Voice Mode. While the older voice model relied on three separate models- one to convert voice to text, another to convert text to voice, and GPT-4 to process prompt. However, the GPT-4o comes with multimodal capabilities and is capable of doing a variety of tasks.
Although advanced Voice Mode is yet to be rolled out for wider ChatGPT Plus users, below are some steps on how to use the innovative feature when it becomes widely available.
In order to start a conversation with the advanced Voice Mode, users would be required to select the voice icon that will soon appear next to the mic icon.
After a user begins a conversation, they will be taken to another screen where they will be able to mute or unmute their microphone by selecting the microphone icon. One can also end the conversation by pressing the red icon on the bottom right.
During the conversation, users can switch between standard Voice Mode and advanced Voice Mode, which can be selected from the top center of the screen.
OpenAI has said that the usage of advanced Voice Mode (audio inputs and outputs) will be limited on a daily basis, and precise limits are subject to change. The ChatGPT app will show a warning when a user is left with three minutes of audio. And, once the limit is reached, the conversation will immediately end following which the users will be prompted to use standard voice mode.
OpenAI has said that the advanced Voice Mode as of now cannot create memories or access previous memories and it also does not have access to custom instructions. As there is no support for memory or custom instructions, conversations with text or standard voice cannot be resumed in advanced Voice Mode.
On irregularities in voice transcripts, OpenAI said that voice conversations with GPT-4o are inherently multimodal, allowing for audio exchange between users and the model. Because of this transcribed audio may not always align with original conversations.
