diff --git a/.circleci/config.yml b/.circleci/config.yml
index fdb57af08f5..b46718858d2 100644
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -80,7 +80,7 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,speech,vision]
+            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,torch-speech,vision]
             - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                 key: v0.4-{{ checksum "setup.py" }}
@@ -116,7 +116,7 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,speech,vision]
+            - run: pip install .[sklearn,tf-cpu,torch,testing,sentencepiece,torch-speech,vision]
             - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                 key: v0.4-{{ checksum "setup.py" }}
@@ -147,7 +147,7 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,flax,torch,testing,sentencepiece,speech,vision]
+            - run: pip install .[sklearn,flax,torch,testing,sentencepiece,torch-speech,vision]
             - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                 key: v0.4-{{ checksum "setup.py" }}
@@ -183,7 +183,7 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,flax,torch,testing,sentencepiece,speech,vision]
+            - run: pip install .[sklearn,flax,torch,testing,sentencepiece,torch-speech,vision]
             - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                 key: v0.4-{{ checksum "setup.py" }}
@@ -213,7 +213,7 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,torch,testing,sentencepiece,speech,vision,timm]
+            - run: pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]
             - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                   key: v0.4-torch-{{ checksum "setup.py" }}
@@ -248,7 +248,7 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,torch,testing,sentencepiece,speech,vision,timm]
+            - run: pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm]
             - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                   key: v0.4-torch-{{ checksum "setup.py" }}
@@ -277,7 +277,7 @@ jobs:
                       - v0.4-tf-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece]
+            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece,tf-speech]
             - save_cache:
                   key: v0.4-tf-{{ checksum "setup.py" }}
                   paths:
@@ -310,7 +310,7 @@ jobs:
                       - v0.4-tf-{{ checksum "setup.py" }}
                       - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece]
+            - run: pip install .[sklearn,tf-cpu,testing,sentencepiece,tf-speech]
             - save_cache:
                   key: v0.4-tf-{{ checksum "setup.py" }}
                   paths:
@@ -338,7 +338,7 @@ jobs:
                     - v0.4-flax-{{ checksum "setup.py" }}
                     - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: sudo pip install .[flax,testing,sentencepiece]
+            - run: sudo pip install .[flax,testing,sentencepiece,flax-speech,vision]
             - save_cache:
                   key: v0.4-flax-{{ checksum "setup.py" }}
                   paths:
@@ -371,7 +371,7 @@ jobs:
                     - v0.4-flax-{{ checksum "setup.py" }}
                     - v0.4-{{ checksum "setup.py" }}
             - run: pip install --upgrade pip
-            - run: sudo pip install .[flax,testing,sentencepiece]
+            - run: sudo pip install .[flax,testing,sentencepiece,vision,flax-speech]
             - save_cache:
                   key: v0.4-flax-{{ checksum "setup.py" }}
                   paths:
@@ -401,7 +401,7 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,torch,testing,sentencepiece,speech,vision]
+            - run: pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision]
             - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                   key: v0.4-torch-{{ checksum "setup.py" }}
@@ -437,7 +437,7 @@ jobs:
                       - v0.4-{{ checksum "setup.py" }}
             - run: sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev
             - run: pip install --upgrade pip
-            - run: pip install .[sklearn,torch,testing,sentencepiece,speech,vision]
+            - run: pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision]
             - run: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cpu.html
             - save_cache:
                   key: v0.4-torch-{{ checksum "setup.py" }}
diff --git a/.github/workflows/self-push.yml b/.github/workflows/self-push.yml
index 3a755f22d12..f112f8b82da 100644
--- a/.github/workflows/self-push.yml
+++ b/.github/workflows/self-push.yml
@@ -33,7 +33,7 @@ jobs:
           apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
           apt install -y libsndfile1-dev
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
 
       - name: Launcher docker
         uses: actions/checkout@v2
@@ -78,6 +78,61 @@ jobs:
           name: run_all_tests_torch_gpu_test_reports
           path: reports
 
+  run_tests_flax_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    container:
+      image: tensorflow/tensorflow:2.4.1-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
+          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
+          pip install --upgrade pip
+          pip install .[sklearn,testing,sentencepiece,flax,flax-speech,vision]
+
+      - name: Launcher docker
+        uses: actions/checkout@v2
+        with:
+          fetch-depth: 2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
+          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
+      
+#      - name: Fetch the tests to run
+#        run: |
+#          python utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
+
+      - name: Report fetched tests
+        uses: actions/upload-artifact@v2
+        with:
+          name: test_fetched
+          path: test_preparation.txt
+
+      - name: Run all non-slow tests on GPU
+        run: |
+          python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_flax_gpu
+#          if [ -f test_list.txt ]; then
+#            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_flax_gpu $(cat test_list.txt)
+#          fi
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        run: cat reports/tests_flax_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_flax_gpu_test_reports
+          path: reports
+
 #  run_tests_tf_gpu:
 #    runs-on: [self-hosted, docker-gpu, single-gpu]
 #    timeout-minutes: 120
@@ -89,7 +144,7 @@ jobs:
 #        run: |
 #          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
 #          pip install --upgrade pip
-#          pip install .[sklearn,testing,onnxruntime,sentencepiece]
+#          pip install .[sklearn,testing,onnxruntime,sentencepiece,tf-speech]
 #
 #      - name: Launcher docker
 #        uses: actions/checkout@v2
@@ -147,7 +202,7 @@ jobs:
           apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
           apt install -y libsndfile1-dev
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+          pip install .[sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
       
       - name: Launcher docker
         uses: actions/checkout@v2
@@ -195,6 +250,61 @@ jobs:
           name: run_all_tests_torch_multi_gpu_test_reports
           path: reports
 
+  run_tests_flax_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    container:
+      image: tensorflow/tensorflow:2.4.1-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Install dependencies
+        run: |
+          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
+          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
+          pip install --upgrade pip
+          pip install .[sklearn,testing,sentencepiece,flax,flax-speech,vision]
+
+      - name: Launcher docker
+        uses: actions/checkout@v2
+        with:
+          fetch-depth: 2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
+          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
+      
+#      - name: Fetch the tests to run
+#        run: |
+#          python utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
+
+      - name: Report fetched tests
+        uses: actions/upload-artifact@v2
+        with:
+          name: test_fetched
+          path: test_preparation.txt
+
+      - name: Run all non-slow tests on GPU
+        run: |
+          python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_flax_multi_gpu
+#          if [ -f test_list.txt ]; then
+#            python -m pytest -n 2 --dist=loadfile -v --make-reports=tests_flax_multi_gpu $(cat test_list.txt)
+#          fi
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        run: cat reports/tests_flax_multi_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_flax_multi_gpu_test_reports
+          path: reports
+
 #  run_tests_tf_multi_gpu:
 #    runs-on: [self-hosted, docker-gpu, multi-gpu]
 #    timeout-minutes: 120
@@ -206,7 +316,7 @@ jobs:
 #        run: |
 #          apt -y update && apt install -y software-properties-common && apt -y update && add-apt-repository -y ppa:git-core/ppa && apt -y update && apt install -y git
 #          pip install --upgrade pip
-#          pip install .[sklearn,testing,onnxruntime,sentencepiece]
+#          pip install .[sklearn,testing,onnxruntime,sentencepiece,tf-speech]
 #
 #      - name: Launcher docker
 #        uses: actions/checkout@v2
diff --git a/.github/workflows/self-scheduled.yml b/.github/workflows/self-scheduled.yml
index a1ad79b4300..834653af9d0 100644
--- a/.github/workflows/self-scheduled.yml
+++ b/.github/workflows/self-scheduled.yml
@@ -34,7 +34,7 @@ jobs:
         run: |
           apt -y update && apt install -y libsndfile1-dev git
           pip install --upgrade pip
-          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -85,6 +85,45 @@ jobs:
           name: run_all_tests_torch_gpu_test_reports
           path: reports
 
+  run_all_tests_flax_gpu:
+    runs-on: [self-hosted, docker-gpu, single-gpu]
+    container:
+      image: tensorflow/tensorflow:2.4.1-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          pip install --upgrade pip
+          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
+          pip install .[flax,integrations,sklearn,testing,sentencepiece,flax-speech,vision]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
+          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_flax_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_flax_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_flax_gpu_test_reports
+          path: reports
+
   run_all_tests_tf_gpu:
     runs-on: [self-hosted, docker-gpu, single-gpu]
     container:
@@ -102,7 +141,7 @@ jobs:
         run: |
           apt -y update && apt install -y git
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnx,sentencepiece]
+          pip install .[sklearn,testing,onnx,sentencepiece,tf-speech]
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -158,7 +197,7 @@ jobs:
         run: |
           apt -y update && apt install -y libsndfile1-dev git
           pip install --upgrade pip
-          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,speech,vision,timm]
+          pip install .[integrations,sklearn,testing,onnxruntime,sentencepiece,torch-speech,vision,timm]
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -213,7 +252,7 @@ jobs:
         run: |
           apt -y update && apt install -y git
           pip install --upgrade pip
-          pip install .[sklearn,testing,onnx,sentencepiece]
+          pip install .[sklearn,testing,onnx,sentencepiece,tf-speech]
 
       - name: Are GPUs recognized by our DL frameworks
         run: |
@@ -251,6 +290,45 @@ jobs:
           name: run_all_tests_tf_multi_gpu_test_reports
           path: reports
 
+  run_all_tests_flax_multi_gpu:
+    runs-on: [self-hosted, docker-gpu, multi-gpu]
+    container:
+      image: tensorflow/tensorflow:2.4.1-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Launcher docker
+        uses: actions/checkout@v2
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Install dependencies
+        run: |
+          pip install --upgrade pip
+          pip install --upgrade "jax[cuda111]" -f https://storage.googleapis.com/jax-releases/jax_releases.html
+          pip install .[flax,integrations,sklearn,testing,sentencepiece,flax-speech,vision]
+
+      - name: Are GPUs recognized by our DL frameworks
+        run: |
+          python -c "from jax.lib import xla_bridge; print('GPU available:', xla_bridge.get_backend().platform)"
+          python -c "import jax; print('Number of GPUs available:', len(jax.local_devices()))"
+
+      - name: Run all tests on GPU
+        run: |
+          python -m pytest -n 1 -v --dist=loadfile --make-reports=tests_flax_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_flax_gpu_failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_flax_gpu_test_reports
+          path: reports
+
   run_all_tests_torch_cuda_extensions_gpu:
     runs-on: [self-hosted, docker-gpu, single-gpu]
     container:
diff --git a/setup.py b/setup.py
index f7e28b2f5d4..a797c163827 100644
--- a/setup.py
+++ b/setup.py
@@ -1,4 +1,4 @@
-# Copyright 2020 The HuggingFace Team. All rights reserved.
+# Copyright 2021 The HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -64,7 +64,6 @@ from pathlib import Path
 
 from setuptools import find_packages, setup
 
-
 # Remove stale transformers.egg-info directory to avoid https://github.com/pypa/pip/issues/5466
 stale_egg_info = Path(__file__).parent / "transformers.egg-info"
 if stale_egg_info.exists():
@@ -253,7 +252,11 @@ extras["ray"] = deps_list("ray[tune]")
 extras["integrations"] = extras["optuna"] + extras["ray"]
 
 extras["serving"] = deps_list("pydantic", "uvicorn", "fastapi", "starlette")
-extras["speech"] = deps_list("soundfile", "torchaudio")
+extras["audio"] = deps_list("soundfile")
+extras["speech"] = deps_list("torchaudio") + extras["audio"]  # `pip install ".[speech]"` is deprecated and `pip install ".[torch-speech]"` should be used instead
+extras["torch-speech"] = deps_list("torchaudio") + extras["audio"]
+extras["tf-speech"] = extras["audio"]
+extras["flax-speech"] = extras["audio"]
 extras["vision"] = deps_list("Pillow")
 extras["timm"] = deps_list("timm")
 extras["codecarbon"] = deps_list("codecarbon")
@@ -275,7 +278,7 @@ extras["all"] = (
     + extras["flax"]
     + extras["sentencepiece"]
     + extras["tokenizers"]
-    + extras["speech"]
+    + extras["torch-speech"]
     + extras["vision"]
     + extras["integrations"]
     + extras["timm"]