Merge pull request #81 from storytold/echelon-test-weights

Add zero shot weights and upload six test files; add to db seed tool

crates/cli/dev_database_seed/src/seeding/zero_shot_tts.rs

@@ -18,9 +18,12 @@ pub async fn seed_zero_shot_tts(mysql_pool: &Pool<MySql>) -> AnyhowResult<()> {
   };
 
   let records = [
-    ("Goku", "todo", &user_token),
-    ("David Attenborough", "todo", &user_token),
-    ("Ash Ketchum", "todo", &user_token),
+    ("Alice", "qtqaprnd5shtybve4fqpvcfp50yjw238fbgj92z1521c50xqdxy1akkhkw7tesj0", &user_token),
+    ("Biden", "n945w0xsq15xrh16hc147a5mc1a91gwh886e14qqzte1gr9z9q3yawjfvanp4fmg", &user_token),
+    ("Goku", "cnnv05yjst2m737dpmxazgfpksjf4y7cxxern2ph7gddgnkh2bw1ephg5mhjbz14", &user_token),
+    ("Hilary", "7wav68ba2yy86491jk36cgk36tkmzesr452dgfs28wchkrr03bd0h3e6c1bbz9eg", &user_token),
+    ("Obama", "z3gy4v56sgtfrxfrpvaj7v74sqc67rcqs89jb884b00zfdm9vmkf1w2fsnta0gwp", &user_token),
+    ("Trump", "qcy7pv3rph0ntkqnpz5cfg9ksyh7kkz53v1wbr2ckvt8znvhxqn7ca5mz7wzm3q5", &user_token),
   ];
 
   for (voice_name, bucket_hash, user_token) in records {

crates/lib/api_clients/cloud_storage/src/bucket_path_unifier.rs

@@ -17,6 +17,8 @@ pub struct BucketPathUnifier {
   pub w2l_inference_output_root: PathBuf,
   pub w2l_model_root: PathBuf,
   pub w2l_end_bump_root: PathBuf,
+  // Zero shot
+  pub zs_tts_speaker_embedding_root: PathBuf,
   // VC
   pub rvc_v2_model_root: PathBuf,
   pub softvc_model_root: PathBuf,
@@ -39,6 +41,8 @@ impl BucketPathUnifier {
       w2l_inference_output_root: PathBuf::from("/w2l_inference_output"),
       w2l_model_root: PathBuf::from("/w2l_pretrained_models"),
       w2l_end_bump_root: PathBuf::from("/w2l_end_bumps"),
+      // Zero shot
+      zs_tts_speaker_embedding_root: PathBuf::from("/user/zero_shot_embeddings/tts"),
       // VC
       rvc_v2_model_root: PathBuf::from("/user_uploaded_rvc_v2_models"),
       softvc_model_root: PathBuf::from("/user_uploaded_softvc_models"),
@@ -197,6 +201,19 @@ impl BucketPathUnifier {
       .join(video_filename)
   }
 
+  // ==================== ZERO SHOT SPEAKER EMBEDDINGS ==================== //
+
+  // NB: Entropic hash is not based on file hash and will be reused for future encodings.
+  pub fn zero_shot_tts_speaker_encoding(&self, entropic_hash: &str, version: u32) -> PathBuf {
+    let hashed_path = Self::hashed_directory_path(entropic_hash);
+    let speaker_encoding_filename = format!("{}_{}.zs", &entropic_hash, version);
+
+    self.zs_tts_speaker_embedding_root
+        .join(hashed_path)
+        .join(speaker_encoding_filename)
+  }
+
+
   // ==================== VOICE CONVERSION MODELS ==================== //
 
   // NB: Entropic hash is not based on file hash and is shared between .pth and .index files.
@@ -279,6 +296,7 @@ mod tests {
       w2l_model_root: PathBuf::from("/test_path_w2l_pretrained_models"),
       w2l_end_bump_root: PathBuf::from("/test_path_w2l_end_bumps"),
       // VOICE CONVERSION
+      zs_tts_speaker_embedding_root: PathBuf::from("/test/zs_tts_speaker_embedding_root"),
       rvc_v2_model_root: PathBuf::from("/test_path_rvc_v2_models"),
       softvc_model_root: PathBuf::from("/test_path_softvc_models"),
       so_vits_svc_model_root: PathBuf::from("/test_path_so_vits_svc_models"),
@@ -383,6 +401,27 @@ mod tests {
                "/test_path_w2l_output/a/b/c/vocodes_video_TYPEABCDEF.mp4");
   }
 
+  #[test]
+  fn test_zero_shot_tts_speaker_encoding_paths() {
+    let paths = get_instance();
+    let entropic_hash = "entropic";
+    let version = 0;
+    assert_eq!(paths.zero_shot_tts_speaker_encoding(entropic_hash, version).to_str().unwrap(),
+               "/test/zs_tts_speaker_embedding_root/e/n/t/entropic_0.zs");
+
+    let version = 23;
+    assert_eq!(paths.zero_shot_tts_speaker_encoding(entropic_hash, version).to_str().unwrap(),
+               "/test/zs_tts_speaker_embedding_root/e/n/t/entropic_23.zs");
+  }
+
+  //#[test]
+  //fn test_temp() {
+  //  // TODO/TEMP: Generating paths to use in dev
+  //  let unifier = BucketPathUnifier::default_paths();
+  //  let path = unifier.zero_shot_tts_speaker_encoding("qcy7pv3rph0ntkqnpz5cfg9ksyh7kkz53v1wbr2ckvt8znvhxqn7ca5mz7wzm3q5", 0);
+  //  assert_eq!(path.to_str().unwrap(), "");
+  //}
+
   #[test]
   fn test_rvc_model_and_index_paths() {
     let paths = get_instance();

crates/service/job/download_job/src/job_types/voice_conversion/rvc_v2/process_rvc_v2_model.rs

@@ -180,3 +180,15 @@ pub async fn process_rvc_v2_model<'a, 'b>(
     entity_type: Some(VoiceConversionModelType::RvcV2.to_string()), // NB: This may be different from `GenericDownloadType` in the future!
   })
 }
+
+//#[cfg(test)]
+//mod tests {
+//  use crockford::crockford_entropy_lower;
+//
+//  #[test]
+//  fn temp() {
+//    // need to make some hashes really quick
+//    let e = crockford_entropy_lower(64);
+//    assert_eq!(e, "");
+//  }
+//}

crates/service/job/inference_job/src/main.rs

@@ -247,6 +247,22 @@ async fn main() -> AnyhowResult<()> {
     job_stats: job_stats.clone(),
   };
 
+  //// TODO/REMOVE - testing download of the dev weights
+  //let records = [
+  //  ("Alice", "qtqaprnd5shtybve4fqpvcfp50yjw238fbgj92z1521c50xqdxy1akkhkw7tesj0"),
+  //  ("Biden", "n945w0xsq15xrh16hc147a5mc1a91gwh886e14qqzte1gr9z9q3yawjfvanp4fmg"),
+  //  ("Goku", "cnnv05yjst2m737dpmxazgfpksjf4y7cxxern2ph7gddgnkh2bw1ephg5mhjbz14"),
+  //  ("Hilary", "7wav68ba2yy86491jk36cgk36tkmzesr452dgfs28wchkrr03bd0h3e6c1bbz9eg"),
+  //  ("Obama", "z3gy4v56sgtfrxfrpvaj7v74sqc67rcqs89jb884b00zfdm9vmkf1w2fsnta0gwp"),
+  //  ("Trump", "qcy7pv3rph0ntkqnpz5cfg9ksyh7kkz53v1wbr2ckvt8znvhxqn7ca5mz7wzm3q5"),
+  //];
+  //let paths = BucketPathUnifier::default_paths();
+  //for (name, hash) in records {
+  //  let object_path = paths.zero_shot_tts_speaker_encoding(hash, 0);
+  //  let filesystem_path = format!("{}_weights.npz", name);
+  //  private_bucket_client.download_file_to_disk(object_path, filesystem_path).await?;
+  //}
+
   let job_dependencies = JobDependencies {
     scoped_execution: ScopedExecution::new_from_env()?,
     fs: FileSystemDetails {