Audio recording guide
High-quality audio data is vital for model training. This topic provides guidelines for creating high-quality recordings, including the recording environment, devices, and procedure.
Recording device
Mobile phone
Recording environment
Environment selection
Select a recording environment that reduces noise and reverberation. Record in a small room that is less than 10 square meters. Rooms with sound-absorbing materials provide the best results. You can also use inexpensive acoustic foam to modify a room. The foam changes sound reflections on flat surfaces into diffuse reflections. This reduces reverberation and improves audio quality.
Noise control
Outdoor noise: Close doors and windows to reduce outdoor noise.
Indoor noise: Common sources of indoor noise include air conditioners, fans (including computer fans), fluorescent light ballasts, and human voices. You can use a mobile phone to record the ambient sound and listen to it at a high volume. This helps you identify and turn off any noise sources.
Reverberation control
Reverberation is the effect that occurs when sound reflects, refracts, and diffuses in a space, causing it to linger and fade gradually. Reflections from smooth surfaces, such as walls and glass, can make the audio sound muddy.
When recording, avoid empty rooms. Choose a location with sound-absorbing features or irregular surfaces to reduce reverberation. Standard offices and conference rooms are not suitable for recording because they often have high reverberation.
Instructions
A typical bedroom is a good place to record. When recording, follow these guidelines:
Stay about 10 cm away from your phone. This distance helps prevent popping sounds and electrical noise.
Close doors and windows to reduce outdoor noise.
Turn off air conditioners and fans to reduce background noise.
Draw the curtains to reduce sound reflection from glass.
Open closet doors. You can also cover hard surfaces, such as cabinets and desks, with clothes, bedsheets, or other soft items. This reduces sound reflection from smooth surfaces and improves the recording quality.
Recording script
Avoid very short sentences in your script. When reading, maintain a smooth and continuous flow and avoid frequent, unnecessary pauses. A pause of 5 seconds or longer can negatively affect the cloning result or even cause it to fail.
Familiarize yourself with the script before you record. Decide on a persona and delivery style. Avoid reading in a monotone. This helps ensure the cloned voice meets your expectations.
There are no special restrictions on the script content. You can use the same text that you plan to synthesize.
If your script contains both Chinese and English text, you only need to record the Chinese parts. The cloned voice will automatically be able to speak both Chinese and English.