diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 3ea6e96..8c25236 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -28,6 +28,8 @@ repos:
     hooks:
     -   additional_dependencies:
         -   types-setuptools
+        -   types-lxml
+        -   numpy  # for numpy plugin
         id: mypy
 
 -   repo: https://gitlab.com/vojko.pribudic/pre-commit-update
diff --git a/pyproject.toml b/pyproject.toml
index ce32d56..05075e7 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -60,9 +60,20 @@ markers = [
 
 
 [tool.mypy]
+plugins = ["numpy.typing.mypy_plugin"]
+
 ignore_missing_imports = true
 
 
+strict = true
+
+disallow_subclassing_any = false
+# ❗ error: Class cannot subclass "Processor" (has type "Any")
+disallow_any_generics = false
+disallow_untyped_defs = false
+disallow_untyped_calls = false
+
+
 [tool.ruff]
 select = ["E", "F", "I"]
 ignore = [
diff --git a/requirements-dev.txt b/requirements-dev.txt
index de6003d..16ae880 100644
--- a/requirements-dev.txt
+++ b/requirements-dev.txt
@@ -7,5 +7,6 @@ ruff
 pytest-ruff
 
 mypy
+types-lxml
 types-setuptools
 pytest-mypy
diff --git a/src/dinglehopper/align.py b/src/dinglehopper/align.py
index c5f12f7..5d1f290 100644
--- a/src/dinglehopper/align.py
+++ b/src/dinglehopper/align.py
@@ -4,8 +4,7 @@ from math import ceil
 from typing import Optional
 
 from rapidfuzz.distance import Levenshtein
-
-from .edit_distance import grapheme_clusters
+from uniseg.graphemecluster import grapheme_clusters
 
 
 def align(t1, t2):
diff --git a/src/dinglehopper/character_error_rate.py b/src/dinglehopper/character_error_rate.py
index 35d3b07..88a88f8 100644
--- a/src/dinglehopper/character_error_rate.py
+++ b/src/dinglehopper/character_error_rate.py
@@ -1,5 +1,5 @@
 import unicodedata
-from typing import List, Tuple
+from typing import List, Tuple, TypeVar
 
 from multimethod import multimethod
 from uniseg.graphemecluster import grapheme_clusters
@@ -7,6 +7,8 @@ from uniseg.graphemecluster import grapheme_clusters
 from .edit_distance import distance
 from .extracted_text import ExtractedText
 
+T = TypeVar("T")
+
 
 @multimethod
 def character_error_rate_n(
@@ -34,21 +36,24 @@ def character_error_rate_n(
 def _(reference: str, compared: str) -> Tuple[float, int]:
     seq1 = list(grapheme_clusters(unicodedata.normalize("NFC", reference)))
     seq2 = list(grapheme_clusters(unicodedata.normalize("NFC", compared)))
-    return character_error_rate_n(seq1, seq2)
+    cer, n = character_error_rate_n(seq1, seq2)
+    return cer, n
 
 
 @character_error_rate_n.register
 def _(reference: ExtractedText, compared: ExtractedText) -> Tuple[float, int]:
-    return character_error_rate_n(
+    cer, n = character_error_rate_n(
         reference.grapheme_clusters, compared.grapheme_clusters
     )
+    return cer, n
 
 
-def character_error_rate(reference, compared) -> float:
+def character_error_rate(reference: T, compared: T) -> float:
     """
     Compute character error rate.
 
     :return: character error rate
     """
+    cer: float
     cer, _ = character_error_rate_n(reference, compared)
     return cer
diff --git a/src/dinglehopper/cli.py b/src/dinglehopper/cli.py
index 5d2000a..a58a2af 100644
--- a/src/dinglehopper/cli.py
+++ b/src/dinglehopper/cli.py
@@ -1,5 +1,6 @@
 import os
 from collections import Counter
+from typing import List
 
 import click
 from jinja2 import Environment, FileSystemLoader
@@ -76,7 +77,7 @@ def gen_diff_report(
         if o is not None:
             o_pos += len(o)
 
-    found_differences = dict(Counter(elem for elem in found_differences))
+    counted_differences = dict(Counter(elem for elem in found_differences))
 
     return (
         """
@@ -87,7 +88,7 @@ def gen_diff_report(
         """.format(
             gtx, ocrx
         ),
-        found_differences,
+        counted_differences,
     )
 
 
@@ -113,7 +114,7 @@ def process(
     metrics: bool = True,
     differences: bool = False,
     textequiv_level: str = "region",
-):
+) -> None:
     """Check OCR result against GT.
 
     The @click decorators change the signature of the decorated functions, so we keep
@@ -122,8 +123,8 @@ def process(
 
     gt_text = extract(gt, textequiv_level=textequiv_level)
     ocr_text = extract(ocr, textequiv_level=textequiv_level)
-    gt_words: list = list(words_normalized(gt_text))
-    ocr_words: list = list(words_normalized(ocr_text))
+    gt_words: List[str] = list(words_normalized(gt_text))
+    ocr_words: List[str] = list(words_normalized(ocr_text))
 
     assert isinstance(gt_text, ExtractedText)
     assert isinstance(ocr_text, ExtractedText)
diff --git a/src/dinglehopper/cli_summarize.py b/src/dinglehopper/cli_summarize.py
index e0c20cb..c49911b 100644
--- a/src/dinglehopper/cli_summarize.py
+++ b/src/dinglehopper/cli_summarize.py
@@ -1,5 +1,6 @@
 import json
 import os
+from typing import Dict
 
 import click
 from jinja2 import Environment, FileSystemLoader
@@ -13,8 +14,8 @@ def process(reports_folder, occurrences_threshold=1):
     wer_list = []
     cer_sum = 0
     wer_sum = 0
-    diff_c = {}
-    diff_w = {}
+    diff_c: Dict[str, int] = {}
+    diff_w: Dict[str, int] = {}
 
     for report in os.listdir(reports_folder):
         if report.endswith(".json"):
diff --git a/src/dinglehopper/edit_distance.py b/src/dinglehopper/edit_distance.py
index ac4a847..ec564ae 100644
--- a/src/dinglehopper/edit_distance.py
+++ b/src/dinglehopper/edit_distance.py
@@ -9,7 +9,7 @@ from .extracted_text import ExtractedText
 
 
 @multimethod
-def distance(seq1: List[str], seq2: List[str]):
+def distance(seq1: List[str], seq2: List[str]) -> int:
     """Compute the Levenshtein edit distance between two lists of grapheme clusters.
 
     This assumes that the grapheme clusters are already normalized.
@@ -20,7 +20,7 @@ def distance(seq1: List[str], seq2: List[str]):
 
 
 @distance.register
-def _(s1: str, s2: str):
+def _(s1: str, s2: str) -> int:
     """Compute the Levenshtein edit distance between two Unicode strings
 
     Note that this is different from levenshtein() as this function knows about Unicode
@@ -33,7 +33,7 @@ def _(s1: str, s2: str):
 
 
 @distance.register
-def _(s1: ExtractedText, s2: ExtractedText):
+def _(s1: ExtractedText, s2: ExtractedText) -> int:
     return Levenshtein.distance(s1.grapheme_clusters, s2.grapheme_clusters)
 
 
diff --git a/src/dinglehopper/extracted_text.py b/src/dinglehopper/extracted_text.py
index af54d7c..e4b0915 100644
--- a/src/dinglehopper/extracted_text.py
+++ b/src/dinglehopper/extracted_text.py
@@ -4,7 +4,7 @@ import re
 import unicodedata
 from contextlib import suppress
 from itertools import repeat
-from typing import List, Optional
+from typing import Any, Dict, List, Optional
 
 import attr
 import numpy as np
@@ -173,10 +173,11 @@ class ExtractedText:
     normalization = attr.ib(converter=Normalization, default=Normalization.NFC_SBB)
 
     @property
-    def text(self):
+    def text(self) -> str:
         if self._text is not None:
             return self._text
         else:
+            assert self.joiner is not None and self.segments is not None
             return self.joiner.join(s.text for s in self.segments)
 
     @functools.cached_property
@@ -186,6 +187,7 @@ class ExtractedText:
         This property is cached.
         """
 
+        assert self.joiner is not None
         if len(self.joiner) > 0:
             joiner_grapheme_cluster = list(grapheme_clusters(self.joiner))
             assert len(joiner_grapheme_cluster) == 1  # see joiner's check above
@@ -203,6 +205,7 @@ class ExtractedText:
         else:
             # TODO Test with text extracted at glyph level (joiner == "")
             clusters = []
+            assert self.segments is not None
             for seg in self.segments:
                 clusters += seg.grapheme_clusters + self._joiner_grapheme_cluster
             clusters = clusters[:-1]
@@ -218,6 +221,7 @@ class ExtractedText:
             else:
                 # Recurse
                 segment_id_for_pos = []
+                assert self.joiner is not None and self.segments is not None
                 for s in self.segments:
                     seg_ids = [s.segment_id_for_pos(i) for i in range(len(s.text))]
                     segment_id_for_pos.extend(seg_ids)
@@ -280,7 +284,7 @@ def invert_dict(d):
     return {v: k for k, v in d.items()}
 
 
-def get_textequiv_unicode(text_segment, nsmap) -> str:
+def get_textequiv_unicode(text_segment: Any, nsmap: Dict[str, str]) -> str:
     """Get the TextEquiv/Unicode text of the given PAGE text element."""
     segment_id = text_segment.attrib["id"]
     textequivs = text_segment.findall("./page:TextEquiv", namespaces=nsmap)
@@ -304,7 +308,7 @@ def get_first_textequiv(textequivs, segment_id):
     if np.any(~nan_mask):
         if np.any(nan_mask):
             log.warning("TextEquiv without index in %s.", segment_id)
-        index = np.nanargmin(indices)
+        index = int(np.nanargmin(indices))
     else:
         # try ordering by conf
         confidences = np.array([get_attr(te, "conf") for te in textequivs], dtype=float)
@@ -313,7 +317,7 @@ def get_first_textequiv(textequivs, segment_id):
                 "No index attributes, use 'conf' attribute to sort TextEquiv in %s.",
                 segment_id,
             )
-            index = np.nanargmax(confidences)
+            index = int(np.nanargmax(confidences))
         else:
             # fallback to first entry in case of neither index or conf present
             log.warning("No index attributes, use first TextEquiv in %s.", segment_id)
@@ -321,7 +325,7 @@ def get_first_textequiv(textequivs, segment_id):
     return textequivs[index]
 
 
-def get_attr(te, attr_name) -> float:
+def get_attr(te: Any, attr_name: str) -> float:
     """Extract the attribute for the given name.
 
     Note: currently only handles numeric values!
diff --git a/src/dinglehopper/ocr_files.py b/src/dinglehopper/ocr_files.py
index be66719..f9bd977 100644
--- a/src/dinglehopper/ocr_files.py
+++ b/src/dinglehopper/ocr_files.py
@@ -1,6 +1,6 @@
 import os
 import sys
-from typing import Iterator
+from typing import Dict, Iterator, Optional
 
 import chardet
 from lxml import etree as ET
@@ -10,11 +10,11 @@ from uniseg.graphemecluster import grapheme_clusters
 from .extracted_text import ExtractedText, normalize_sbb
 
 
-def alto_namespace(tree: ET.ElementTree) -> str:
+def alto_namespace(tree: ET._ElementTree) -> Optional[str]:
     """Return the ALTO namespace used in the given ElementTree.
 
     This relies on the assumption that, in any given ALTO file, the root element has the
-    local name "alto". We do not check if the files uses any valid ALTO namespace.
+    local name "alto". We do not check if the file uses any valid ALTO namespace.
     """
     root_name = ET.QName(tree.getroot().tag)
     if root_name.localname == "alto":
@@ -23,8 +23,15 @@ def alto_namespace(tree: ET.ElementTree) -> str:
         raise ValueError("Not an ALTO tree")
 
 
-def alto_extract_lines(tree: ET.ElementTree) -> Iterator[ExtractedText]:
-    nsmap = {"alto": alto_namespace(tree)}
+def alto_nsmap(tree: ET._ElementTree) -> Dict[str, str]:
+    alto_ns = alto_namespace(tree)
+    if alto_ns is None:
+        raise ValueError("Could not determine ALTO namespace")
+    return {"alto": alto_ns}
+
+
+def alto_extract_lines(tree: ET._ElementTree) -> Iterator[ExtractedText]:
+    nsmap = alto_nsmap(tree)
     for line in tree.iterfind(".//alto:TextLine", namespaces=nsmap):
         line_id = line.attrib.get("ID")
         line_text = " ".join(
@@ -37,7 +44,7 @@ def alto_extract_lines(tree: ET.ElementTree) -> Iterator[ExtractedText]:
         # FIXME hardcoded SBB normalization
 
 
-def alto_extract(tree: ET.ElementTree) -> ExtractedText:
+def alto_extract(tree: ET._ElementTree) -> ExtractedText:
     """Extract text from the given ALTO ElementTree."""
     return ExtractedText(None, list(alto_extract_lines(tree)), "\n", None, None)
 
@@ -98,7 +105,7 @@ def extract_texts_from_reading_order_group(group, tree, nsmap, textequiv_level):
     if ET.QName(group.tag).localname in ["OrderedGroup", "OrderedGroupIndexed"]:
         ro_children = list(group)
 
-        ro_children = filter(lambda child: "index" in child.attrib.keys(), ro_children)
+        ro_children = [child for child in ro_children if "index" in child.attrib.keys()]
         ro_children = sorted(ro_children, key=lambda child: int(child.attrib["index"]))
     elif ET.QName(group.tag).localname in ["UnorderedGroup", "UnorderedGroupIndexed"]:
         ro_children = list(group)
diff --git a/src/dinglehopper/word_error_rate.py b/src/dinglehopper/word_error_rate.py
index afb4fe0..b6e0a3a 100644
--- a/src/dinglehopper/word_error_rate.py
+++ b/src/dinglehopper/word_error_rate.py
@@ -1,5 +1,5 @@
 import unicodedata
-from typing import Iterable, Tuple
+from typing import Generator, Iterable, Tuple, TypeVar
 
 import uniseg.wordbreak
 from multimethod import multimethod
@@ -7,6 +7,8 @@ from rapidfuzz.distance import Levenshtein
 
 from .extracted_text import ExtractedText
 
+T = TypeVar("T")
+
 # Did we patch uniseg.wordbreak.word_break already?
 word_break_patched = False
 
@@ -32,7 +34,7 @@ def patch_word_break():
 
 
 @multimethod
-def words(s: str):
+def words(s: str) -> Generator[str, None, None]:
     """Extract words from a string"""
 
     global word_break_patched
@@ -61,34 +63,36 @@ def words(s: str):
 
 
 @words.register
-def _(s: ExtractedText):
-    return words(s.text)
+def _(s: ExtractedText) -> Generator[str, None, None]:
+    yield from words(s.text)
 
 
 @multimethod
-def words_normalized(s: str):
-    return words(unicodedata.normalize("NFC", s))
+def words_normalized(s: str) -> Generator[str, None, None]:
+    yield from words(unicodedata.normalize("NFC", s))
 
 
 @words_normalized.register
-def _(s: ExtractedText):
-    return words_normalized(s.text)
+def _(s: ExtractedText) -> Generator[str, None, None]:
+    yield from words_normalized(s.text)
 
 
 @multimethod
 def word_error_rate_n(reference: str, compared: str) -> Tuple[float, int]:
     reference_seq = list(words_normalized(reference))
     compared_seq = list(words_normalized(compared))
-    return word_error_rate_n(reference_seq, compared_seq)
+    wer, n = word_error_rate_n(reference_seq, compared_seq)
+    return wer, n
 
 
 @word_error_rate_n.register
 def _(reference: ExtractedText, compared: ExtractedText) -> Tuple[float, int]:
-    return word_error_rate_n(reference.text, compared.text)
+    wer, n = word_error_rate_n(reference.text, compared.text)
+    return wer, n
 
 
 @word_error_rate_n.register
-def _(reference: Iterable, compared: Iterable) -> Tuple[float, int]:
+def _(reference: Iterable[T], compared: Iterable[T]) -> Tuple[float, int]:
     reference_seq = list(reference)
     compared_seq = list(compared)
 
@@ -102,6 +106,7 @@ def _(reference: Iterable, compared: Iterable) -> Tuple[float, int]:
     return d / n, n
 
 
-def word_error_rate(reference, compared) -> float:
+def word_error_rate(reference: T, compared: T) -> float:
+    wer: float
     wer, _ = word_error_rate_n(reference, compared)
     return wer