tacotron2

crates/service/job/inference_job/src/job/job_types/tts/tacotron2_v2_early_fakeyou/mod.rs

@@ -1,4 +1,5 @@
 pub mod process_job;
 pub mod read_metadata_file;
 pub mod seconds_to_decoder_steps;
+pub mod tacotron2_dependencies;
 pub mod tacotron2_inference_command;

crates/service/job/inference_job/src/job/job_types/tts/tacotron2_v2_early_fakeyou/process_job.rs

@@ -32,7 +32,6 @@ use crate::util::maybe_download_file_from_bucket::maybe_download_file_from_bucke
 /// This allows the request to bypass the model cache and query the latest TTS model.
 const TEST_REQUEST_TEXT: &str = "This is a test request.";
 
-
 pub struct ProcessJobArgs<'a> {
   pub job_dependencies: &'a JobDependencies,
   pub job: &'a AvailableInferenceJob,
@@ -50,10 +49,18 @@ pub async fn process_job(args: ProcessJobArgs<'_>) -> Result<JobSuccessResult, P
       .new_generic_inference(job.inference_job_token.as_str())
       .map_err(|e| ProcessSingleJobError::Other(anyhow!(e)))?;
 
+  let model_dependencies = args
+      .job_dependencies
+      .job_specific_dependencies
+      .maybe_tacotron2_dependencies
+      .as_ref()
+      .ok_or_else(|| ProcessSingleJobError::JobSystemMisconfiguration(Some("missing Tacotron2 dependencies".to_string())))?;
+
   // ==================== CONFIRM OR DOWNLOAD WAVEGLOW VOCODER MODEL ==================== //
 
   let waveglow_vocoder_model_fs_path = {
-    let waveglow_vocoder_model_filename = args.job_dependencies.job_type_details.tacotron2_old_vocodes.waveglow_vocoder_model_filename.clone();
+    // TODO(bt,2023-11-21): Port this to the common downloader code.
+    let waveglow_vocoder_model_filename = model_dependencies.waveglow_vocoder_model_filename.clone();
     let waveglow_vocoder_model_fs_path = args.job_dependencies.fs.semi_persistent_cache.tts_pretrained_vocoder_model_path(&waveglow_vocoder_model_filename);
     let waveglow_vocoder_model_object_path = args.job_dependencies.bucket_path_unifier.tts_pretrained_vocoders_path(&waveglow_vocoder_model_filename);
 
@@ -74,7 +81,8 @@ pub async fn process_job(args: ProcessJobArgs<'_>) -> Result<JobSuccessResult, P
   // ==================== CONFIRM OR DOWNLOAD HIFIGAN (NORMAL) VOCODER MODEL ==================== //
 
   let pretrained_hifigan_vocoder_model_fs_path = {
-    let hifigan_vocoder_model_filename = args.job_dependencies.job_type_details.tacotron2_old_vocodes.hifigan_vocoder_model_filename.clone();
+    // TODO(bt,2023-11-21): Port this to the common downloader code.
+    let hifigan_vocoder_model_filename = model_dependencies.hifigan_vocoder_model_filename.clone();
     let hifigan_vocoder_model_fs_path = args.job_dependencies.fs.semi_persistent_cache.tts_pretrained_vocoder_model_path(&hifigan_vocoder_model_filename);
     let hifigan_vocoder_model_object_path = args.job_dependencies.bucket_path_unifier.tts_pretrained_vocoders_path(&hifigan_vocoder_model_filename);
 
@@ -95,7 +103,8 @@ pub async fn process_job(args: ProcessJobArgs<'_>) -> Result<JobSuccessResult, P
   // ==================== CONFIRM OR DOWNLOAD HIFIGAN (SUPERRES) VOCODER MODEL ==================== //
 
   let hifigan_superres_vocoder_model_fs_path = {
-    let hifigan_superres_vocoder_model_filename = args.job_dependencies.job_type_details.tacotron2_old_vocodes.hifigan_superres_vocoder_model_filename.clone();
+    // TODO(bt,2023-11-21): Port this to the common downloader code.
+    let hifigan_superres_vocoder_model_filename = model_dependencies.hifigan_superres_vocoder_model_filename.clone();
     let hifigan_superres_vocoder_model_fs_path = args.job_dependencies.fs.semi_persistent_cache.tts_pretrained_vocoder_model_path(&hifigan_superres_vocoder_model_filename);
     let hifigan_superres_vocoder_model_object_path = args.job_dependencies.bucket_path_unifier.tts_pretrained_vocoders_path(&hifigan_superres_vocoder_model_filename);
 
@@ -113,7 +122,7 @@ pub async fn process_job(args: ProcessJobArgs<'_>) -> Result<JobSuccessResult, P
     hifigan_superres_vocoder_model_fs_path
   };
 
-//  // ==================== CONFIRM OR DOWNLOAD OPTIONAL CUSTOM VOCODER MODEL ==================== //
+  // ==================== CONFIRM OR DOWNLOAD OPTIONAL CUSTOM VOCODER MODEL ==================== //
 
   let custom_vocoder_fs_path = match &tts_model.maybe_custom_vocoder {
     None => None,
@@ -249,7 +258,7 @@ pub async fn process_job(args: ProcessJobArgs<'_>) -> Result<JobSuccessResult, P
 
   let inference_start_time = Instant::now();
 
-  let _r = args.job_dependencies.job_type_details.tacotron2_old_vocodes.inference_command.execute_inference(InferenceArgs {
+  let _r = model_dependencies.inference_command.execute_inference(InferenceArgs {
     synthesizer_checkpoint_path: &tts_synthesizer_fs_path,
     text_pipeline_type: text_pipeline_type_or_guess.to_str(),
     vocoder: vocoder_option,

crates/service/job/inference_job/src/job/job_types/tts/tacotron2_v2_early_fakeyou/tacotron2_dependencies.rs

@@ -0,0 +1,36 @@
+use errors::AnyhowResult;
+use crate::job::job_types::tts::tacotron2_v2_early_fakeyou::tacotron2_inference_command::Tacotron2InferenceCommand;
+
+pub struct Tacotron2Dependencies {
+  pub inference_command: Tacotron2InferenceCommand,
+
+  /// Common pretrained waveglow vocoder filename
+  pub waveglow_vocoder_model_filename: String,
+
+  /// Common pretrained hifigan vocoder filename
+  pub hifigan_vocoder_model_filename: String,
+
+  /// Common pretrained hifigan super resolution vocoder filename
+  pub hifigan_superres_vocoder_model_filename: String,
+}
+
+impl Tacotron2Dependencies {
+  pub fn setup() -> AnyhowResult<Self> {
+
+    let waveglow_vocoder_model_filename = easyenv::get_env_string_or_default(
+      "TTS_WAVEGLOW_VOCODER_MODEL_FILENAME", "waveglow.pth");
+
+    let hifigan_vocoder_model_filename = easyenv::get_env_string_or_default(
+      "TTS_HIFIGAN_VOCODER_MODEL_FILENAME", "hifigan.pth");
+
+    let hifigan_superres_vocoder_model_filename = easyenv::get_env_string_or_default(
+      "TTS_HIFIGAN_SUPERRES_VOCODER_MODEL_FILENAME", "hifigan_superres.pth");
+
+    Ok(Self {
+      inference_command: Tacotron2InferenceCommand::from_env()?,
+      waveglow_vocoder_model_filename,
+      hifigan_vocoder_model_filename,
+      hifigan_superres_vocoder_model_filename,
+    })
+  }
+}

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

@@ -24,7 +24,6 @@ use mysql_queries::queries::tts::tts_models::get_tts_model_for_inference_improve
 use mysql_queries::queries::voice_conversion::inference::get_voice_conversion_model_for_inference::VoiceConversionModelForInference;
 use newrelic_telemetry::Client as NewRelicClient;
 
-use crate::job::job_types::tts::tacotron2_v2_early_fakeyou::tacotron2_inference_command::Tacotron2InferenceCommand;
 use crate::job::job_types::tts::vall_e_x::model_downloaders::VallEXDownloaders;
 use crate::job::job_types::tts::vall_e_x::vall_e_x_inference_command::VallEXCreateEmbeddingCommand;
 use crate::job::job_types::tts::vall_e_x::vall_e_x_inference_command::VallEXInferenceCommand;
@@ -158,25 +157,8 @@ pub struct JobCaches {
 
 /// Per-job type details
 pub struct JobTypeDetails {
-  pub tacotron2_old_vocodes: Tacotron2VocodesDetails,
   pub vits: VitsDetails,
   pub vall_e_x:  VallEXDetails
-  //pub tacotron2_modern: ...,
-  //pub softvc: ...,
-}
-
-/// "Old" TT2 (vocodes-era)
-pub struct Tacotron2VocodesDetails {
-  pub inference_command: Tacotron2InferenceCommand,
-
-  /// Common pretrained waveglow vocoder filename
-  pub waveglow_vocoder_model_filename: String,
-
-  /// Common pretrained hifigan vocoder filename
-  pub hifigan_vocoder_model_filename: String,
-
-  /// Common pretrained hifigan super resolution vocoder filename
-  pub hifigan_superres_vocoder_model_filename: String,
 }
 
 pub struct VitsDetails {

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

@@ -4,6 +4,7 @@ use enums::by_table::generic_inference_jobs::inference_model_type::InferenceMode
 use errors::AnyhowResult;
 
 use crate::job::job_types::lipsync::sad_talker::sad_talker_dependencies::SadTalkerDependencies;
+use crate::job::job_types::tts::tacotron2_v2_early_fakeyou::tacotron2_dependencies::Tacotron2Dependencies;
 use crate::job::job_types::vc::rvc_v2::rvc_v2_dependencies::RvcV2Dependencies;
 use crate::job::job_types::vc::so_vits_svc::svc_dependencies::SvcDependencies;
 use crate::util::scoped_execution::ScopedExecution;
@@ -12,6 +13,7 @@ pub struct JobSpecificDependencies {
   pub maybe_rvc_v2_dependencies: Option<RvcV2Dependencies>,
   pub maybe_sad_talker_dependencies: Option<SadTalkerDependencies>,
   pub maybe_svc_dependencies: Option<SvcDependencies>,
+  pub maybe_tacotron2_dependencies: Option<Tacotron2Dependencies>,
 }
 
 impl JobSpecificDependencies {
@@ -20,6 +22,7 @@ impl JobSpecificDependencies {
     let mut maybe_rvc_v2_dependencies = None;
     let mut maybe_sad_talker_dependencies = None;
     let mut maybe_svc_dependencies = None;
+    let mut maybe_tacotron2_dependencies = None;
 
     if scoped_execution.can_run_job(InferenceModelType::RvcV2) {
       info!("Setting RVCv2 dependencies...");
@@ -36,10 +39,16 @@ impl JobSpecificDependencies {
       maybe_svc_dependencies = Some(SvcDependencies::setup()?);
     }
 
+    if scoped_execution.can_run_job(InferenceModelType::Tacotron2) {
+      info!("Setting Tacotron2 dependencies...");
+      maybe_tacotron2_dependencies = Some(Tacotron2Dependencies::setup()?);
+    }
+
     Ok(JobSpecificDependencies {
       maybe_rvc_v2_dependencies,
       maybe_sad_talker_dependencies,
-      maybe_svc_dependencies
+      maybe_svc_dependencies,
+      maybe_tacotron2_dependencies,
     })
   }
 }

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

@@ -50,12 +50,11 @@ use subprocess_common::docker_options::{DockerEnvVar, DockerFilesystemMount, Doc
 use crate::http_server::run_http_server::CreateServerArgs;
 use crate::http_server::run_http_server::launch_http_server;
 use crate::job::job_loop::main_loop::main_loop;
-use crate::job::job_types::tts::tacotron2_v2_early_fakeyou::tacotron2_inference_command::Tacotron2InferenceCommand;
 use crate::job::job_types::tts::vall_e_x::model_downloaders::VallEXDownloaders;
 use crate::job::job_types::tts::vall_e_x::vall_e_x_inference_command::VallEXCreateEmbeddingCommand;
 use crate::job::job_types::tts::vall_e_x::vall_e_x_inference_command::VallEXInferenceCommand;
 use crate::job::job_types::tts::vits::vits_inference_command::VitsInferenceCommand;
-use crate::job_dependencies::{FileSystemDetails, JobCaches, JobDependencies, JobTypeDetails, JobWorkerDetails, Tacotron2VocodesDetails, VallEXDetails, VitsDetails};
+use crate::job_dependencies::{FileSystemDetails, JobCaches, JobDependencies, JobTypeDetails, JobWorkerDetails, VallEXDetails, VitsDetails};
 use crate::job_specific_dependencies::JobSpecificDependencies;
 use crate::util::scoped_execution::ScopedExecution;
 use crate::util::scoped_temp_dir_creator::ScopedTempDirCreator;
@@ -169,15 +168,6 @@ async fn main() -> AnyhowResult<()> {
 
   let common_env = CommonEnv::read_from_env()?;
 
-  let waveglow_vocoder_model_filename = easyenv::get_env_string_or_default(
-    "TTS_WAVEGLOW_VOCODER_MODEL_FILENAME", "waveglow.pth");
-
-  let hifigan_vocoder_model_filename = easyenv::get_env_string_or_default(
-    "TTS_HIFIGAN_VOCODER_MODEL_FILENAME", "hifigan.pth");
-
-  let hifigan_superres_vocoder_model_filename = easyenv::get_env_string_or_default(
-    "TTS_HIFIGAN_SUPERRES_VOCODER_MODEL_FILENAME", "hifigan_superres.pth");
-
   let sidecar_max_synthesizer_models = easyenv::get_env_num(
     "SIDECAR_MAX_SYNTHESIZER_MODELS", 3)?;
 
@@ -305,13 +295,7 @@ async fn main() -> AnyhowResult<()> {
     low_priority_starvation_prevention_every_nth,
     maybe_minimum_priority,
     job_type_details: JobTypeDetails {
-      tacotron2_old_vocodes: Tacotron2VocodesDetails {
-        inference_command: Tacotron2InferenceCommand::from_env()?,
-        waveglow_vocoder_model_filename,
-        hifigan_vocoder_model_filename,
-        hifigan_superres_vocoder_model_filename,
-      },
-      vall_e_x: VallEXDetails { 
+      vall_e_x: VallEXDetails {
         downloaders: VallEXDownloaders::build_all_from_env(),
         inference_command: VallEXInferenceCommand::from_env()?, 
         create_embedding_command: VallEXCreateEmbeddingCommand::from_env()?,