# Before Start : > - Well, this time (unlike the TV ad I thought was the world's first with a voice-over, it was just a fictitious representation by a fiverr person), the basic audio recording used as a dataset is indeed the oldest we could find with a human voice. > - As you'd expect, this model isn't meant to be perfect. I've tried to create a "clean" version, but this simply distorts and renders useless the template, as it's simply not indicative of the original sample. Created **Just for fun**. > - The voice you might hear would be that of Édouard-Léon Scott de Martinville, but we're not sure. Indeed, given that the "recording" is made from a Phonautograph, it is not theoretically possible to find the exact pitch of the voice. The model is therefore based on assumptions made by experts in the restoration field, which has brought almost everyone into agreement today. > - As you can imagine, the image on the thumbnail is not a real photo. It's a technique I used with ia GFPGAN 1.4 to make it look like a photo, so that it could create a "real life" face from a simple period drawing. Which worked well. Then I just searched for period clothing to integrate his face. # Last Update : <t:1703786764:R> > - **Model URL (1860 epochs just for fun) :** > - All Previews here use this version ! > - https://huggingface.co/rayzox57/1860_Au_Clair_De_La_Lune_Correct_Pitch_RVC/resolve/main/1860_Au_Clair_De_La_Lune_Correct_Pitch_v2_1860e.zip > - **Language:** > - Not really important, but it's French > - **Version :** > - <a:firev2:1167361499149381662> RVC V2.0 > - **Pitch Extraction Algorithm :** > - RMVPE > - **Epochs - Steps :** > - 1.86k - 5.55k > - 900 - 2.7k > - **Dataset :** > - ~ 00:00:21 (Original Sound with Correct Pitch) > - **Recommended Usage :** > - All > - **Search Feature Ratio :** > - 0.75 > - **Pitch :** > - Logic Pitch ( -4 = Man / -12 = Women ) > - You can adjust if you found an better result # Previews : > - **Preview_TTS.wav :** > - *Provide* : French TTS > - *Contains External Effects* : No > - *Pitch* : -4 > - *Feature Ratio* : 0.75 > > - **Preview_Cover.wav :** > - *Provide* : An Cover of "Au clair de la lune", the same sing you can hear on original > - *Contains External Effects* : Yes > - *Pitch* : -4 > - *Feature Ratio* : 0.75

References : > - **1840_Au_Clair_De_La_Lune_Unclean_Version** : the version i use for the dataset, possible correct pitch > - **1840_Au_Clair_De_La_Lune_Clean_Version** : The best result I could get by cleaning the audio (the pitch on this one isn't the same, though).