Replay: The EASIEST way to create AI Cover Songs!

Bob Doyle Media
26 Mar 202419:03

TLDRIn this video, the host introduces Replay, a free AI tool that allows users to change the voice in any song to any voice they choose. The process is straightforward: users can upload a song, select a voice model from the Weights G platform, and Replay will separate the vocals, convert them, and allow users to download the new version. The host demonstrates converting a song created in Sunno with different voice models, including Dean Martin and Squidward from SpongeBob. The tool also offers features like multimodel, which combines multiple voice models for a song, and the ability to create music from a text prompt. The host emphasizes the fun and creative potential of Replay, encouraging viewers to experiment with it.

Takeaways

  • 🎶 Replay is a free tool for voice conversion in songs, allowing users to change any song's voice to a voice of their choice.
  • 🔍 The software is available for Windows, Mac, and Linux, and it operates locally on the user's machine.
  • 📚 To get started, users need to download Replay and create an account on Weights & Biases (W&B) to access voice models.
  • 🎵 Users can upload their own music tracks, record a vocal track, or input a YouTube URL for Replay to process the audio.
  • 🔗 The voice models can be searched and downloaded from W&B, with a wide variety of singers and characters available.
  • 📉 The tool allows for adjustments such as relative pitch to match the original song's vocal range to the new voice.
  • ⏱️ The conversion process can be time-consuming, depending on the user's hardware, particularly the GPU used.
  • 🔄 Users can download individual tracks (vocals, instrumentals) for remixing purposes.
  • 🎧 There is an option to use multiple voice models for a single song, known as the multimodel feature.
  • 🧪 The software also has a feature to merge different voice models to create a new, unique voice.
  • ✂️ Post-conversion editing can be done using a multitrack editor for further refinement and to remove any noise from the vocal track.
  • 📝 The tool can also convert speech to different voices, not just music.

Q & A

  • What is the name of the AI tool discussed in the video that allows users to create AI cover songs?

    -The AI tool discussed in the video is called 'Replay'.

  • Is Replay a free tool to use?

    -Yes, Replay is a free tool and does not require any subscriptions to use its features.

  • What platforms is Replay available for?

    -Replay is available for Windows, Mac, and Linux.

  • How does Replay handle the process of changing the voice in a song?

    -Replay allows users to upload a song, select a voice model, and then it separates the audio, converts the vocals, and merges them back together with the instrumental track.

  • What is the name of the website where users can download additional voice models for Replay?

    -The website for downloading additional voice models is weights.ai.

  • How does one search for specific voice models on weights.ai?

    -On weights.ai, users can create an account and use the search function by typing the name of the voice or character they are looking for.

  • What file types are used by Replay to represent voice models?

    -Replay uses '.pth' and '.index' files to represent voice models.

  • What is the purpose of the 'relative pitch' setting in Replay?

    -The 'relative pitch' setting in Replay is used to adjust the pitch of the voice model to match the original song's vocal range.

  • Can Replay be used to convert speech as well as music?

    -Yes, Replay can also be used to convert speech by using the pre-stemmed option and selecting a voice model.

  • What is the 'multimodel' feature in Replay?

    -The 'multimodel' feature in Replay allows users to select multiple voice models to use for one song, creating a batch of converted tracks with different voices.

  • How long does it typically take for Replay to separate and convert a song once the process has started?

    -The initial audio separation can take a couple of minutes, but subsequent conversions are faster, especially if using a powerful GPU.

  • What is the process for creating music from a text prompt using Replay?

    -Users can input a text prompt describing the style and instrument of the desired music. Replay then uses models to generate a short snippet of music based on that description.

Outlines

00:00

🎙️ Voice Cloning and Music Conversion with Replay

The video introduces Replay, a free tool for voice conversion that allows users to take any song and replace its vocals with a voice of their choice. The host discusses their experience using Replay to convert songs over the weekend and emphasizes the tool's ease of use. The process includes downloading the Replay software, selecting a song or recording a vocal track, choosing a voice model from the Weights website, and customizing pitch to match the original song's vocal range. The host also mentions the option to create songs from text prompts, although this feature is considered less practical compared to other tools like Sunno.

05:01

🔄 Customizing and Converting Vocals with Replay

This paragraph demonstrates how to customize the vocal conversion process in Replay. It covers the settings for voice conversion, such as stem only, pre-stemmed, relative pitch, and instrument pitch. The host shows how to audition different voices, including those of famous singers and characters, and how to adjust the pitch to match the original vocals. The video also includes a live demonstration of converting a song's vocals to those of Dean Martin and Billy Eillish, highlighting the ability to download individual tracks for further remixing.

10:02

🎼 Advanced Features of Replay: Multimodel and Text-to-Music

The host explores advanced features of Replay, including the ability to use multiple voice models for a single song and to merge different voice models to create a new vocal style. The paragraph also touches on the process of converting speech to different voices, such as changing the host's voice to sound like Squidward from SpongeBob. Additionally, the video briefly mentions the text-to-music feature, which generates short audio snippets based on text prompts, but notes that this feature may be less useful due to the time-consuming process and the availability of higher-quality alternatives like Sunno.

15:03

🚀 Final Thoughts on Replay and Creative Applications

The video concludes with the host's enthusiasm for Replay and its potential for creative applications. They encourage viewers to experiment with the tool, create duets, and remix songs using any singer's voice. The host also compares Replay's text-to-music feature to Sunno's capabilities, demonstrating Sunno's superior output. The video ends with a call to action for viewers interested in AI and creativity to subscribe to the channel for more content on similar topics.

Mindmap

Keywords

💡Replay

Replay is a software tool for voice conversion, allowing users to take any song and change its voice to any other voice they choose. It is central to the video's theme as the host demonstrates how to use Replay to create AI cover songs by changing the original vocals to different voices, like those of Dean Martin or Squidward from SpongeBob.

💡Voice Cloning

Voice cloning refers to the process of replicating a person's voice, which is a significant aspect covered in the video. The host discusses the use of Replay for voice cloning, not in a 100% total conversion sense, but enough to make the process of changing a song's voice very accessible and easy.

💡AI Cover Songs

AI cover songs are songs that use artificial intelligence to replicate or modify the original vocals. The video is focused on creating such covers using Replay, showcasing the ease with which one can transform songs by swapping out the original vocals with AI-generated ones.

💡Weights and Biases (Weights)

Weights and Biases, often abbreviated as 'Weights', is a platform hosting over 20,000 voice models that can be used with Replay. The host guides viewers on how to create an account, search for, and download voice models to use within Replay for voice conversion.

💡Model Download

A model download refers to the process of obtaining voice models from Weights and Biases to use in Replay. These models are essential for voice conversion, and the host explains how to download and implement them for creating AI cover songs.

💡Track Separation

Track separation is the process of separating the vocal track from the instrumental track of a song. Replay automates this process, which is vital for replacing the original vocals with AI-generated ones, as shown in the video.

💡Relative Pitch

Relative pitch is a feature in Replay that allows users to adjust the pitch of the voice model to match the original song's key. The host uses this feature when changing a song's voice to a higher-pitched model, such as Billy Eillish's, to ensure the converted voice fits the music.

💡Multimodel

Multimodel is a feature in Replay that enables users to select multiple voice models for a single song, creating a batch of voice-converted songs. The host demonstrates this by converting a song with both Darth Vader and Garth Brooks' voices.

💡Merge Models

Merge models is an advanced feature of Replay that combines two different voice models to create an entirely new voice. The host merges Billy Eillish's and Sheldon Plankton's voices to create a unique vocal sound for a song.

💡Pre-stemmed

Pre-stemmed refers to audio that has already had its vocals separated from the music. The host mentions using the pre-stemmed option when recording one's own voice, as there is no need for the software to separate vocals in this case.

💡Sunno

Sunno is a music creation tool that is mentioned in the context of generating an original song for the purpose of the demonstration. The host uses Sunno to create a song that is then used to showcase the capabilities of Replay in converting and manipulating vocals.

Highlights

Replay is a tool for voice conversion, allowing users to change any song's voice to a voice of their choice.

The Replay project is entirely free, with no subscriptions required.

Replay is available for Windows, Mac, and Linux platforms.

The tool can download models from the internet, which are used for voice conversion.

Users can generate their own songs or convert existing ones, including from YouTube URLs.

Replay offers over 20,000 voice models to choose from, including singers and characters.

The voice conversion process is done locally on the user's machine.

Users can adjust the relative pitch to match the original song's vocal range.

Replay allows for the creation of songs with multiple voice models, offering a variety of vocal styles.

The tool can also be used to convert speech, not just music.

Replay provides the ability to create music from a text prompt, although this feature is considered less practical.

The quality of separated vocal tracks is highly praised for its clarity.

Users can audition different voice models quickly once the initial audio separation is done.

Replay enables users to create duets and remix songs with downloaded individual tracks.

The tool can merge models to create an entirely new voice, offering unique vocal combinations.

Replay is described as being highly addictive and fun to use for converting and creating music.

The host spent hours using Replay to create humorous voice conversions from YouTube videos.

Replay is positioned as a powerful tool for those interested in AI and creative music production.