ChatGPT is a powerful language model, capable of generating remarkably human-like text.
However, when those responses are read aloud using text-to-speech (TTS), they can sometimes have an artificial or robotic quality.
If you aim for shockingly natural spoken output, it’s time to add some spice to the mix! Let’s look at techniques to get that authentic speech feel.
How to improve your ChatGPT output
In everyday conversation, people rarely speak in perfectly formed, grammatically correct sentences. We use filler words like “um,” “uh,” “like,” and “you know”. These might seem like verbal clutter, but they actually serve an important function in natural speech.
You can instruct ChatGPT’s voice feature to include these fillers in its responses strategically. This will help it sound more conversational.
Here’s an example:
- Prompt: Explain the difference between weather and climate.
- Tweaked prompt: Could you explain, um, the difference between weather and climate? Maybe use some ‘likes’ and ‘you knows’ too.
TTS systems will convert those filler words into pauses and hesitations, making the output feel more organic. Experiment to find the right level for your desired style!
Here is the original response:
And here is the response with our tweaked prompt:
Besides that, people naturally use contractions in everyday speech. “Cannot” becomes “can’t,” “it is” becomes “it’s,” and so on. These make speech flow more smoothly and feel conversational. You can encourage ChatGPT to use contractions to add to its “human-like” speech pattern.
Real human speech has variation in tone. We use changes in pitch, volume, and speed to stress words or add emotion. Here’s where some targeted punctuation can guide your TTS:
- Adding commas for pauses: “The weather today is sunny, warm, and perfect for a hike.”
- Exclamation points for excitement: “That movie was amazing! I loved the ending.”
- Question marks convey inquisitiveness: “I’ve always wondered, how do birds learn to fly?”
Practice makes progress
Like any skill, making ChatGPT’s spoken output sound more natural takes practice. Try out different combinations of techniques. Pay attention to your favorite podcasts or audiobooks to see how professional speakers use vocal inflections and filler words to great effect.
Remember, there’s no single “right” way to do this!
A note on text-to-speech systems
The quality of your TTS system plays a big role too. Some are better at interpreting punctuation for a lifelike pronunciation than others.
Popular options include:
When might less be more?
While natural-sounding ChatGPT output is excellent in many cases, there are times when a more formal or ‘robotic’ delivery is appropriate. Think about these scenarios:
- Delivering news or factual information: Conciseness and clarity matter here.
- Accessibility: Some users may find too many fillers or tonal changes confusing.
It’s all about finding the best fit for your desired outcome!
Featured image credit: Jason Leung/Unsplash
