5 Ways To Make Money Using OpenAI Whisper (+Bonus Trick)

OpenAI whisper is a State-Of-The-Art (SOTA) speech recognition technology developed by OpenAI. It is free and open source, unlike other SOTAs such as Azure, Google, or Meta, which cost money.

Furthermore, OpenAI is far more accurate and straightforward to use.

Many of us are still unaware of how capable AIs can be in creating things that normal humans cannot even imagine.

Without the proper knowledge, we blame AI for taking over human jobs and work, but as humanity evolves, so will technology. It’s inevitable.

As a result, we must treat AI as a friend rather than a competitor to accomplish our goals.

OpenAI also has other services like Dall-E 2 and ChatGPT. However, people continue to use it for entertainment rather than for income.

We will go over various ways to use OpenAI whisper for both personal and professional purposes.

What is OpenAI Whisper?

According to OpenAI, Whisper is a web-based Automatic Speech Recognition (ASR) system trained on 680,000 hours of multilingual and multitask supervised data.

Whisper demonstrates that using a large and diverse dataset improves robustness to accents, background noise, and technical language.

It also supports transcription in multiple languages and translation from those languages into English.

Whisper is releasing models and inference code as a foundation for developing useful applications and furthering research into robust speech processing.

is OpenAI Whisper accurate?

Compared to other State-Of-The-Art Technologies such as Kaldi, Vosk, wav2vec 2.0, and others, OpenAI whisper had a comparatively Low WER (Word-Error-Count).

Whisper is exceptional, given that it is free and open source. Let’s look at the comparison between different ASR systems.

In long-form transcription, Whisper competes with cutting-edge commercial and open-source ASR systems.

The distribution of word error rates from six ASR systems on seven long-form datasets with input lengths ranging from a few minutes to a few hours is compared.

The boxes represent the quartiles of per-example WERs, with the aggregate WERs for each dataset annotated on each box.

On all datasets, the Whisper model outperforms the best open-source model (NVIDIA STT) and, in most cases, commercial ASR systems.

The catch is that while these results are impressive, there is a different story when it comes to non-English languages.

The graph below depicts the word error rate for each supported language. The lowest Word Error Rate is 3% in Spanish, and the highest is 47% in Nepali.

Below we see the distribution of languages as a function of the Word Error Rate. Of all the 82 languages, 50 of them have Word-Error-Rates greater than 20%.

There are five model sizes available, four of which are English-only, with speed and accuracy tradeoffs. The names of the available models, as well as their approximate memory requirements and relative speed, are listed below.

So, to answer our main question, how can we make money with OpenAI Whisper?

1. Start a Transcription or Translation Service

There are numerous Transcription and Translation Services available, such as gotranscript, transcribeme, rev, and happyscribe.

You can start your own business offering a Top Notch Service for a lot less cost and with much more precision than other services.

This service could be provided to individuals or businesses who require audio or video transcription or translation.

Many podcasters nowadays talk for more than an hour, sometimes even three hours, and as you know, many transcription services charge per minute. As a result, you can offer your services at extremely competitive rates.

Creating a transcription service for medical professionals, such as transcribing patient notes or converting audio recordings of medical procedures into written form, is a good place to start.

Create a transcription service that is specifically tailored to educators’ needs, such as transcribing lectures or converting educational videos into written form.

Transcription services tailored specifically to the needs of legal professionals, such as transcribing court proceedings or converting legal documents to audio format.

All these services could be provided for free or as part of a subscription with in-app purchases.

You can focus on a single niche or offer services to multiple niches in a single package.

2. Create a voice-to-text App

This concept entails creating a mobile app that allows users to take voice notes, which are then transcribed into written text.

You could create a tool that allows users to create and manage tasks by speaking, which is then transcribed into written text.

You could create a mobile app that enables users to convert spoken language into written text.

You could create a mobile app that allows users to send and receive text messages by using their voice, which is then transcribed into written text.

3. Become a Freelancer on Fiverr

You can begin your gig on a website such as Fiverr. As a new member, you must begin offering services at the lowest possible price.

You can get a head start by asking a friend or family member to purchase your services. Treat your family member or friend as if they were a real client, and speak and act professionally.

Right now, Fiverr is overcrowded. To achieve the best results, you must continue to post more and more gigs on a daily basis.

When you look at the top Fiverr freelancers, you’ll notice that the majority of them have more than 50 gigs. So, the more gigs you have, the more likely it is that you will be found.

You can have gigs for YouTubers, Musicians, and Podcasters. You can offer them translation, transcription, or even subtitles of their content. Assist musicians with lyrics. It is freelancing rather than starting a business.

4. Build a Language Learning App

This can be a hit-or-miss idea. Whisper could be used to create a language learning tool that listens to a user’s spoken or written language and gives feedback based on their input.

You can create your own language exchange app that pairs native speakers of two different languages who want to learn the mother tongue of their learning partner.

You can begin by providing limited translations and feedback, followed by more detailed information in the paid plan. You can offer translation and writing exercises based on user interaction.

Individuals who want to learn a new language could be offered this tool as a subscription service.

5. Start your own Free Transcription and Language Detection Website

You can create your own free website that will transcribe recorded audio and convert it to text format. You can also include a language detection feature on that website. Just as huggingface did here.

Is OpenAI Whisper safe to use?

According to their GitHub Repo, it states.

So this is Completely Safe and Free to use.

Problems with OpenAI Whisper

After testing Whisper in multiple projects and languages, there are a few issues that you will undoubtedly encounter in your workflow.

I discovered that Whisper sometimes struggles to understand what is being said.

For example, if you speak in a low-pitched voice and there is background noise behind the dialogue. OpenAI simply ignores the sentence, issuing no errors or warnings.

In some cases, if a video contains multiple languages, Whisper may skip the entire language and only process a single language within the video.

Bonus Trick

As previously discussed, Whisper has some problems. We can reduce it by nearly 90% by doing just one thing.

Use the free Adobe Enhance feature to improve the audio by removing background noise and increasing the file’s lower-pitch audio.

Adobe Enhance currently only accepts 1GB file size and 1 hour of audio. So, if you have a three-hour audio file, you can split it into three separate files, then merge them into a single mp3 or Wav file and use whisper on it.

Conclusion

Whisper is the most accurate open source Automatic Speech Recognition (ASR) system out there. No matter how good AI is, we must keep one thing in mind. AI will never be able to achieve what humans can. Humans will always be superior in the long run.

Whisper will be a valuable tool for both researchers and hackers due to its accuracy and ease of use when compared to other open-source options.

Whisper’s performance is influenced by its compute intensity, so applications that require larger, more powerful versions of Whisper should run it on GPU, whether locally or in the cloud.

You can find more information about OpenAI Whisper here.